International Journal of Intelligent Engineering and Systems

12

Authors: Umar Umar

This research discusses the optimization of capacity and locations for multi single-phase distributed generation (DG) in unbalanced distribution systems to reduce power losses, improve bus voltage while maintaining voltage imbalances and harmonics within the acceptable limits. The optimization of single-phase DG uses new enhanced symbiotic organism search (NeSOS) based genetic algorithm (GA) operator. NeSOS is developed by modifying the mutualism and commensalism phases using the random weighted inverse vector. The parasitism phase is modified by introducing sub-phase in parasitism phase. NeSOS based GA operator is validated using the IEEE 33-bus system and tested using an unbalanced 25-bus distribution system. Validation using the IEEE 33-bus distribution system shows that NeSOS based GA operator produces the lowest power loss compared to, chaotic moth-flame optimization (CMFO) based power loss sensitivity factor (PLSF), particle swarm optimization (PSO), improved analytical (IA), loss sensitivity factor (LSF), and hybrid PSO method. Simulations using an unbalanced 25-bus distribution system indicate that NeSOS and SOS successfully reduces active and reactive power loss by 68.24% and 64.93% respectively, however in terms of convergence rate, NeSOS is, on average, 52.39% faster than SOS.

41

Authors: Inas Jawad Kadhim, Prashan Premaratne, Zena Ahmed Alwan

Watermarking digital media content could undoubtedly discourage and prevent such illegal sharing activities, safeguarding the copyrights of the original creators. Such content could make use of invisible watermarking, not just for robustness and security, but also for imperceptibility. The proposed research presents a frame selection strategy based on the difference between the absolute histograms of consecutive frames. To increase security and robustness and avoid embedding the watermark frame by frame, ten frames with the maximum absolute histogram value are selected for watermark embedding. The size of the watermark would determine the number of non-overlapping blocks that a key frame is divided into in order to embed the watermark. The watermarks are embedded into the LSB of the frame-blocks, which have the most matched location where the embedding error will be minimum with high imperceptibility. Security can be further enhanced by generating a secret key based on non-sequential video frames, block location, and watermark size. The proposed method surpasses other comparable methods in terms of imperceptibility and robustness to image processing and geometrical attacks, and it is also computationally inexpensive. The results of the proposed method outperformed the best state-of-the-art methods in MPSNR and MNCC by 16 dB and 0.018, respectively.

65

Authors: Mohammed Jabbar Mohammed Ameen, Alaa Imran Al-Muttairi, Hussam Jawad Kadhim

This paper proposes a novel medical image watermarking scheme for a multiuser wireless system based on data normalization and polar code. The watermark image is normalized, converted into a set of coefficients using Discrete Cosine Transform (DCT), and then embedded with host medical image coefficients. The watermark image is encrypted using a Baker map and then normalized. The normalized value is converted to binary data, then XORed with a logistic map sequence, and employed as frozen bits in polar code to achieve security, robustness, and imperceptibility over the Non-Orthogonal Multiple Access-Universal Filtered Multicarrier (NOMA-UFMC) transmission scheme. A collection of medical images was employed to evaluate the proposed technique for two users, and the findings are Peak Signal to Noise Ratio (PSNR=69.5761), structural similarity index (SSIM=0.9966), mean square error (MSE=0.4399) and Normalized Cross-Correlation (NCC=0.9997) which demonstrate that it is capable of achieving a high level of imperceptibility and robustness against attacks. In addition, the proposed scheme produced good Bit Error Rate (BER) performance by exploiting the properties of polar code. Compared with the traditional techniques, the proposed approach has higher image quality, a secure watermark image, performs high robustness against various attacks, and has low execution time (about 0.3284 seconds). The suggested method is, therefore, a promising one for protecting medical images in healthcare settings.

92

Authors: E. Suganthi, F. Kurus Malai Selvi

The resilient concept of cloud computing (CC) enables individuals and businesses to access the necessary services following their requirements. The methodology presents various functionalities like storage solutions, deployment platforms, convenient admittance to online services, and more. A significant issue in CC is load balancing (LB), which reduces the performance and effectiveness of the resources. The LB can be attained through (Task / Virtual Machine) scheduling and allocation. This research considers virtual machine (VM) allocation, which allocates VM to a suitable physical machine (PM). Implementing an efficient VM allocation approach is essential for mitigating energy consumption (EC) and service level agreement (SLA) breaches. This study presents an effectual LB model for CC based on state and randomization. Based on host utilization, the proposed approach first identifies the PM’s current state (high, medium, and low). Next, select the suitable PM for VM allocation using a randomization approach. The Cloudsim toolkit is employed for simulating the suggested technique, and the PlanetLab workload is used to evaluate the performance regarding EC and SLA violations. The proposed approach is compared with MFPED (Medium-Fit Power Efficient Decreasing) and HVMAP (Hybrid VM Allocation and Placement). The experimental outcome shows that the proposed technique significantly lowers EC by 40.81% and 39.76% and SLA violation by 96.81% and 95.58% for MFPED and HVMAP methods.

116

Authors: Nour Zawawi, Nermin Negied

The race to halt Alzheimer's disease (AD) in its tracks demands an early warning system. By predicting which mild cognitive impairment (MCI) patients are likely to decline into AD, clinicians can intervene while the window of opportunity remains open. But how to separate the MCI patients bound for AD from those with more benign forms of impairment? The key lies in examining the factors that influence disease progression. While prior studies have scratched the surface, a comprehensive analysis has proven elusive. Enter the Alzheimer's Disease Neuroimaging Initiative database, which tracks AD progression through a wealth of patient characteristics. Leveraging these rich data, our hybrid approach combines survival analysis with machine learning to generate dynamic predictions of time to AD onset. Rather than merely detecting AD early or diagnosing its current state, our model gazes into the future, forecasting progression from MCI to AD before the disease fully erupts. Among similar efforts, the proposed approach stands apart in scale and accuracy, validated on more patients and with higher predictive power than earlier attempts. Even cognitive tests or brain scans alone can foretell decline, with the proposed work achieving a remarkable C-index of 0.85 when evaluated using the whole ADNI dataset not only a sample from it. By revealing who is likely to convert to AD and when, this work enables clinicians to intervene at the critical junction where MCI transitions to inevitable decline. The future of AD treatment may hinge on such early warnings.

139

Authors: V. Sai Geetha Lakshmi, M. Vanithasri, M. Venu Gopal Rao

The multi-objective economic load dispatch problem (ELDP) with non-smooth cost functions and ramp-rate limits presents a challenging optimization task in power systems. This paper proposes the use of a modified cheetah optimizer (MCO) that incorporates opposition-based learning (OBL) and a dynamic adaptive weighting factor to efficiently solve this problem. The simulations are conducted on standard test systems using MATLAB programming. A comparative study is performed, evaluating the performance of the MCO against basic CO and other similar heuristics. The results demonstrate the effectiveness of the MCO in achieving optimal solutions for the multi-objective ELDP with non-smooth cost functions and ramp-rate limits. The proposed approach offers a promising solution for addressing the complex optimization requirements in power system operation and planning. The optimal cost is determined using multi-criterion optimization (MCO) in 3-bus system as $6,838.6434/h, $7,738.789/h, and $8,252.033/h for 700 MW, 800 MW and 850 MW demand levels respectively. The optimal cost is evaluated as $17,988.96/h for the 13-bus test system considering a total demand of 1800 MW. The optimal cost is evaluated as 121960.30 $/hr, for the 40-bus test system considering a total demand of 10500 MW. These outcomes demonstrate the efficacy of MCO in resolving the ELDP with generator and valve controls.

159

Authors: Anand M, Babu S

Research on real-time Facial Expression Recognition (FER) and human emotion prediction has been ongoing. The desire of humans to enable cohabitation between humans as well as machines can only be realized if a machine is able to comprehend, display, and respond to a human’s emotional state. If an Artificial Intelligence (AI) makes judgements on the basis of emotions rather than just algorithms and logic, it will be able to really breathe life without air. Although there have been several attempts to date to do this, it is still believed to be in the early stages of development. Hence, this paper plans to perform the FER prediction with the help of novel intelligent deep learning technology. Initially, the data is gathered from two standard benchmark sources such as FER 2013 and EMOTIC dataset. Next, the pre-processing of the gathered images is accomplished by face detection, rotation rectification, and Local Binary Pattern (LBP). From the pre-processed images, the feature extraction is performed by the Centralized Binary Pattern (CBP) method. These extracted features undergo the final prediction phase that is done by the novel Enhanced Gated Recurrent Unit (EGRU), in which the parameters of GRU are tuned by the nature inspired optimization algorithm referred as Kookaburra Optimization Algorithm (KOA) with the consideration of error minimization as the major objective function. This novel EGRU-KOA predicts the final outcome with consideration of various facial expressions such as sad, surprise, disgust, contempt, happy, and fear respectively. The novelty of the proposed model is demonstrated by comparing it with various conventional models in terms of distinct analysis respectively. The proposed EGRU-KOA in terms of accuracy is 2.78%, 5.60%, 1.63%, and 4.28% better than DBN+QPSO, GDP, CNN, and WKELM respectively. Similarly, the proposed EGRU-KOA with respect to MSE is 84.82%, 87.41%, 88.51%, and 86.82% better than DBN+QPSO, GDP, CNN, and WKELM respectively.

189

Authors: Visalaxi G, Muthukumaravel A

Resource allocation plays a pivotal role in Cloud Computing (CC), significantly impacting its overall performance. In order to fairly spread workload among servers, network ports, and hard drives, task scheduling is a critical module of CC. Cloud computing experiences request overloading as a result of dynamic computing over the internet. To address this challenge, a novel Cloud based Load balancing using Quantum artificial bee colony Optimization Algorithm has been proposed for task scheduling that can significantly enhance the effectiveness of cloud computing operations. The technique focuses on two important aspects of cloud computing: resource allocation and task scheduling. By achieving load balancing between servers, the proposed CLUQOA (Cloud based Load balancing Using Quantum artificial bee colony Optimization Algorithm) improves reliability and lowers expenses, latency, and response times. To show how effective it is, key performance parameters including makespan, resource usage, task migration, task execution time, and response time are assessed. The results of comparative comparisons with other approaches, such as HUNTER (Holistic resoUrce maNagemenT technique for Energy-efficient cloud computing using aRtificial intelligence), FPNSO (Flower Pollination based Non-dominated Sorting Optimization), and ProHPA (Proactive Hybrid Pod Autoscaling), demonstrate the superiority of CLUQOA in terms of improving resource utilization (35.6%, 26.4%, and 13.9%, respectively) and reducing makespan (11.02%, 9.6%, and 10.4%, respectively).

214

Authors: Muhammad Arif Faizin, Dias Tri Kurniasari, Nazhifah Elqolby, Muhammad Aidiel Rachman Putra, Tohari Ahmad

Information and communication technology is growing rapidly, making it the target of various attacks. The attacks can be in the form of data theft, phishing, and Denial of Service (DoS). There are many ways to handle attacks on communication networks, including developing an Intrusion Detection System (IDS) model. Research on IDS has developed a lot and focuses on certain things such as feature selection, dealing with data imbalance problems. Feature selection is essential to the IDS model because of the dataset’s characteristics, which have many features. Besides, the number of features included in the classification can affect the detection performance of the IDS model. This research proposes an IDS combining mutual information with thresholding feature selection and XGBoost classification algorithm. Mutual information is used to measure the dependency between every input feature and the target features. After the amount of information is obtained with mutual information, thresholding is used to decide the best number of features in the classification process. Then, the data are classified using XGBoost selected features. The proposed method was tested using four metrics: accuracy, precision, recall, and f1-score. This study used UNSW-NB15 as the primary dataset to analyze the best combinations of feature selection method and thresholding value. In addition, the proposed method has also been tested using NSL-KDD and CIC-IDS2017 datasets to evaluate the performance compared with previous research. The proposed method performs best using the CIC-IDS2017 dataset with 99.89 % accuracy and 99.68 % F1 score. Furthermore, it can reduce computational training time compared with other IDS methods that only use feature selection or tree-model-based algorithms without thresholds.

239

Authors: Nallasivan. G, Karpagam. T, Geetha. M, Sankarasubramanian. R. S, Kannan. R, Bhuvanesh. A, Poojitha. G

The management of various multimedia assets, such as photos, videos, audio files, and other rich media content, within a cloud computing environment is referred to as managing multimedia resources in the cloud. To suit the needs of applications and users, this entails the effective storage, retrieval, processing, and distribution of multimedia resources. Given the significance of work planning and managing resources in the cloud computing environment, we present a unique hybrid algorithm in this research. Many cloud-based computing systems have made extensive use of traditional scheduling techniques like ant colony optimization (ACO), first come first serve, etc. The cloud gets client tasks at a high rate, so it is important to handle resource allocation for these tasks carefully. Using the improved pelican optimization algorithm, we efficiently distribute the tasks to the virtual machines in this proposed work. The proposed hybrid algorithm (Improved POA + Improved GJO) is then used to distribute and manage the resources (Memory and CPU) as needed by the tasks. According to experimental findings, the accuracy of the proposed technique increases by 1.12%, 2.11%, and 14.2%, respectively. It shows that the proposed method has good accuracy compared with the existing HUNTER, FT-ERM, and RU-VMM approaches.

262

Authors: Abdulaziz Alashaikh

Conditional Preference Networks (CP-nets) have emerged as a prominent and intuitive model for representing condi-tional preferences within the AI community. However, the literature lacks a concrete investigation on their applica-bility within Multiobjective Evolutionary Algorithms (MOEAs). Many optimization problems that are tackled by MOEA have a room for conditional preferences on the decision space and thus reasoning about CP-nets in MOEA is of great importance. In this work, we motivate the use of CP-nets within the MOEA framework and provide a sim-ple yet powerful approach that can be augmented during the search to return solutions that are not only optimized for the objectives but also preferred by the decision maker’s CPnet. Furthermore, we consider the case where the de-cision maker has no clear idea about her preferences at the beginning of the search but willing to engage during the search to answer simple comparison questions. Our experimental results show that it is possible to return solutions that are both optimal and preferred even when the search starts with empty preference information. The obtained solution space achieved at least 23% improvement in the average penalty score.

290

Authors: Gunawan, Ivan Hosea 1 Esther Irawati Setiawan, Kimiya Fujisawa

This research evaluates the performance of End-to-End Neural Coreference Resolution models in English and Indonesian linguistic contexts, drawing particular attention to the model by K. Lee, recognized for its simplified preprocessing methodology. In this research, we use raw text to find and link mentions in documents, and handle different languages. For the evaluation metrics, the English model achieved F1-Scores of 67.94% and 67.14% on the OntoNotes-5.0 development and training sets. The Indonesian model, prepared using a CoNLL-2012 formatted dataset, attained an F1-Score of 68.88% on a 25% segment of the Book of Markus. We also analyzed the additional features integrated into the model to assess their contributions to performance improvement. The findings indicate that while the English model demonstrates generalizability across various coreference challenges, the Indonesian model’s performance is more domain-specific, being particularly effective within the confines of the Book of Mark.

329

Authors: Weerachai Khamwiangsa, Sakon Klongboonjit

All vehicles are exported globally through sea freight, which involves extended storage periods and exposure to high temperatures. Consequently, issues with the brake system have been identified, arising from the presence of air and contaminants within the system. The aim is to minimize the number of Air in the Vehicle Brake System, targeting zero defects while eliminating repair and labor costs. Research integrating TOPSIS and EDAS with FAHP to address design and transportation issues in vehicles has not yet been identified. Through a systematic literature review, specialist questionnaires, and the Fuzzy Analytic Hierarchy Process (FAHP), the study identified the main contributing factors to the brake system issues and established their importance ranking. Based on this analysis, a multi-criteria framework for weights was determined using the FAHP comprehensive evaluation method. Subsequently, the Technique for Order of Preference by Similarity to the Ideal Solution (TOPSIS) method and The Evaluation based on Distance from Average Solution (EDAS) were employed to evaluate and classify the alternatives based on the obtained weights. By specifying a scoring guideline, the attributes of the brake system could be quantified, ensuring objectivity. Consequently, the brake system issue is implemented, verified, and analysed using TOPSIS and EDAS supported by FAHP. the priority ranking of 4 criteria levels and alternatives are calculated with Pi = 0.995, 0.690, and 0.574 respectively in the TOPSIS technique and ASi = 0.796, 0.500, and 0.478 respectively in the EDAS technique. The total Rejection per thousand of Air in a Vehicle Brake System is 0.99 for CA1, 0.12 for CA2, and 0.03 for CA3. The results highlight the effectiveness of integrating TOPSIS and EDAS with FAHP in reducing the presence of Air in the Vehicle Brake System and minimizing repair costs, with a primary focus on enhancing vehicle safety control as the most critical concern. Additionally, the research prioritizes assessing the risk associated with Air in the Vehicle Brake System.

352

Authors: Venkata Vidyalakshmi Guggilam, Gopikrishnan Sundaram

In the relam of the Internet of Things (IoT), prevalence of missing data due to continuous data collection by smart devices necessitates the essential preliminary step of data imputation before engaging in information mining activities. IoT data exhibit robust interconnections in both spatial and temporal dimensions, surpassing the limitations of Euclidean space. Yet, prevailing machine learning and deep learning approaches often focus solely on temporal attributes or capture spatial features exclusively within a Euclidean framework. To address these challenges, this paper introduces a novel network named ST-Bi-LSTM-VAE (Spatio-Temporal Bidirectional Long Short-Term Memory based Variational Auto-Encoder). The architecture of ST-Bi-LSTM-VAE is primarily grounded in the Variational Auto-Encoder (VAE) framework. This innovative approach incorporates two distinct types of VAEs. The first type is dedicated to computing the adjacent matrix of the device network, a crucial input for the Graph Convolutional Network (GCN) essential in capturing intricate spatial relationships among devices. The second type of VAE is specifically tailored for data imputation, leveraging both global spatial and temporal dependencies. Empirical experiments conducted on diverse publicly available datasets substantiate the efficacy of ST-Bi-LSTM-VAE. The results obtained consistently demonstrate that proposed method surpasses baseline techniques in maintaining pattern, structure, and trend across datasets even at 50% missing gap for imputation task with 4.91% performance improvement in case of Intel Berkley Research Laboratory (IBRL) dataset and 3.5% on PRSA dataset.

380

Authors: Sarmad Ibrahim, Akram Jaddoa Khalaf, Shamam Alwash

This paper proposes an integrated method to simultaneously determine the optimal placement and sizing of battery energy storage systems (BESSs) in power distribution networks using the hunger games search algorithm (HGSA). The objective of the proposed method focuses on concurrently reducing power losses and improving the voltage deviation index to improve the distribution system performance while keeping all constraints within permissible limits. The HGSA is a nature-inspired algorithm that simulates the behavior of prey and predators in finding food. In this paper, the HGSA is used to search for the optimal solution in a multi-dimensional search space consisting of the candidate BESS locations and sizes. The proposed method is applied to modified IEEE 69-bus and IEEE 85-bus test distribution systems with five different scenarios, and the results indicate that the HGSA can efficiently determine the optimal placement and sizing of BESSs, resulting in significant power loss reduction (i.e., 69.13-98.08% for the first test system and 52.97-95.03% for the second test system) and voltage deviation improvement (i.e., 94.75-99.89% for the first test system and 12.12-93.37% for the second test system) compared to the base case. Furthermore, the performance of the HGSA is compared with other well-known metaheuristic optimization algorithms (i.e., whale optimization algorithm, chaotic neural network algorithm, genetic algorithm, grey wolf optimizer, and water cycle algorithm), and the results show that the HGSA outperforms these optimization algorithms in attaining the best optimal solution with faster convergence and less number of iterations.

401

Authors: Karrar Sahib Nassrullah, Ivan Viktorovich Stepanyan, Haider Sahib Nasrallah, Neder Jair Mendez Florez, Abdelrrahmane Mohamed Zidoun, Shahad Raouf Mohammed

The focus of this study is a relatively new development in the field of control: machine learning control. This study offers a mathematical framework for machine learning control and explores three symbolic regression-based techniques for supervised and unsupervised learning. One of the challenges associated with machine learning control pertains to the control general synthesis. This entails figuring out a control function contingent upon the object’s state, guaranteeing the attainment of the control objective while optimizing the quality criterion value across all possible initial states within a permissible zone where finding a satisfactory solution occurs within the space of codes. The implementation of the small variations principle within the basic solution is suggested as a viable technique for developing the algorithms of search. This paper extensively discusses three symbolic regression techniques, including Cartesian genetic programming (CGP), synthesized genetic programming (SGP) and parse-matrix evolution (PME). Notably, synthesized genetic programming, being a novel technique, and PME get utilized for the first time to address the general synthesis of control problems. The mathematical expression’s SGP code is a six-row integer matrix; the first row of the matrix represents the functions that take two arguments, while the second and fourth rows represent the functions that take one argument. The third and fifth rows represent the arguments of the mathematical expression, and the sixth row represents the priority. The computational example demonstrates the potential of symbolic regression approaches as unsupervised machine learning control techniques for addressing the machine learning control challenge of general synthesis of control in order to achieve the stability of a mobile robot system. Likewise, practical experience shows that synthesized genetic programming has faster efficiency than Cartesian genetic programming and parse-matrix evolution in discovering solutions, about 2.33 and 2.11 times on average, respectively.

428

Authors: Mays Kareem Jabbar, Thaar A. Kareem

Wireless Sensor Networks (WSNs) play a pivotal role in remote monitoring, surveillance, and Internet of Things (IoT) applications. The efficient utilization of battery-powered sensor nodes in WSNs, given their limited power capacity, is crucial for successful data transmission. Conventional clustering algorithms while efficient in clustering, often lacks the efficient management of data generated by sensor nodes, leading to redundant data in applications like IoT leading to reduced network lifetime. To overcome this issue, this paper introduces a novel approach, named CCOA-DC (Improved Coati Optimization with Cognitive Factor (CCOA) through Data Compression (DC)), in clustering heterogeneous aggregated WSN data. The research unfolds in two novel phases. Initially, a non-negative matrix factorization (NMF) model is introduced data compression for clustering, addressing the challenge of data transmission and energy efficiency. Subsequently, the performance is enhanced through load balancing, featuring dynamic cluster head selection via Improved Coati Optimization (COA) with cognitive factor (C), denoted as CCOA. A distinctive aspect is the incorporation of the NMF data compression technique in both clustering and cluster head selection processes, introducing an energy-efficient, load-balanced, and compressed data aggregation mechanism. The proposed CCOA-DC undergoes rigorous testing, comparing its performance against existing models to validate its superiority. Comparative analyses with renowned models such as TCBDGA, HEED, and FEEC-IIR underscore the distinct advantages of CCOA-DC. Notably, it achieves a reduction of 78.57% of packet loss ratio compared to FEEC-IIR model. The model achieves high packet delivery ratio which is 98.67%, and shows optimized energy consumption of 68.01% Joules. This novel compression-based metaheuristic data aggregation algorithm showcases its effectiveness in addressing the energy conservation challenge, affirming its prominence in the area of WSNs based IoT applications.

460

Authors: Suhas Kamshetty Chinnababu, Anandababu Jayachandra

Globally, diabetes mellitus is the most dangerous disease and it is important to predict disease at an early stage to treat the disease. The learning-based algorithms play a significant part in supporting decision-making in diabetes prediction as well as diagnosis. Machine Learning (ML) based diabetes classification suffers from poor performance due to constraints such as limited labeled data and the challenge of data imbalance. Therefore, this research proposes the Self-adaptive Evolutionary Algorithm (SAEA) with Convolutional Neural Network (CNN) for the prediction and classification of diabetes. For validating the proposed method’s effectiveness, the data is collected from benchmark datasets such as the Pima Indian Diabetes dataset (PIDD), Frankfurt Hospital, Germany, and North California State University (NCSU). Then, the min-max normalization is used for normalizing the data in the pre-processing step. The Chi-square-based feature selection technique is used for selecting the best feature for categorial features and finally, the SAEA with CNN is used for classification and it classifies the disease into diabetes and non-diabetes. The proposed method’s effectiveness is estimated by using various matrices and it attains the accuracy of 99.87% and 99.99% by using PIDD and German datasets when compared to existing approaches like Deep Neural network (DNN), CNN, and Deep CNN (DCNN).

485

Authors: Dandy Pramana Hostiadi, Tohari Ahmad, Muhammad Aidiel Rachman Putra, Gede Angga Pradipta, Putu Desiana Wulaning Ayu, Made Liandana

The number of cases in the cyber era has increased significantly, which are caused by malicious software known as malware. This malicious software penetrates the network, infects several computers, and forms a collection of zombie computer networks commonly known as Botnets. These botnet threats can gravely impact valuable system resources and stored data and cause severe financial losses if not handled appropriately. Several previous studies introduced a botnet detection model using algorithms from machine learning by optimizing the feature selection process and having high detection results. However, feature selection is carried out without determining the role of features in the mandatory and non-mandatory categories. In fact, not all features can be selected because they have an important role and influence detection performance. This paper proposes a detection model by optimizing feature selection techniques. The initial process is to categorize features into mandatory and non-mandatory features. The feature selection process is carried out on non-mandatory features using two approaches: Univariate and ANOVA. Then, the best features from the feature selection results are aggregated with the Mandatory features and processed in a classification model for detecting malware attacks. The aim is to obtain the best features used in the classification model to improve detection performance by measuring accuracy, precision, and recall. The classification model used is a Decision tree and was tested on three different datasets, namely CTU-13, NCC, and NCC-2. The experiment result obtained an accuracy of 99.27% on the CTU-13 dataset, 98.96% on the NCC dataset, and 98.87% on the NCC-2 dataset. The resulting average precision value is 98.68% in the CTU-13 dataset, 98.26% in the NCC dataset, and 97.90% in the NCC-2 dataset. Finally, the resulting average recall value was 99.27% on the CTU-13 dataset, 98.96% on the NCC dataset, and 98.87% on the NCC-2 dataset. The detection results showed better results than previous research. This model can make analyzing attacks easier and determine treatment when a malware attack occurs.

515

Authors: Shivashanker K, Janaki M

The process of designing and controlling modular multilevel converter (MMC) poses numerous challenges. So, hardware-in-the-loop (HIL) simulation and transient analysis are necessary to test the efficacy and reliability of MMC in different operational conditions. This paper presents transient analysis of MMC using MATLAB/Simulink, and results are verified using HIL simulation to validate the performance of decoupled current and circulating current controllers in real-time (RT). The decoupled current control enables the independent control of real and reactive power in AC system, and circulating current control minimizes the circulating currents in MMC phase-legs. The transient analysis is performed on the study system with a rating of 1000 MVA, 220 kV in MATLAB/Simulink. The analysis is carried out for a step perturbation of 100 MW in the real power and 2.0 kA in reactive current references, and results show the satisfactory responses with MMC controllers. The simulation results are validated with HIL simulation of MMC using two RT simulators. The HIL simulation results show the satisfactory responses with closely matching responses to MATLAB/Simulink results.

539

Authors: Annapoorani S, Lakshmi M

Depression is a widespread psychological disorder that significantly impacts individuals. The conventional diagnostic approach through doctor-patient communication lacks objectivity and precision. The development of a more accurate and balanced technique for the detection of depression is critical. Due to its ability to effectively describe brain activity, resting electroencephalography (rEEG) is a valuable technique for investigating brain differences between healthy controls and patients with depression. This study examined the rEEG data from 95 healthy controls and 119 individuals with depression. This study constructed the brain's functional network and extracted linear spectral power density parameters from the rEEG data using correlation. Extracting distinctive features for identifying depression in rEEG data remains challenging due to the absence of crucial multichannel and temporal contexts in most existing methods. This research proposes taking advantage of an attention-based convolutional neural network with Long Short-Time Memory (ACNN-LSTM) to enhance its ability to identify of depressive disorders by extracting more selective features from rEEG signals. The ACNN-LSTM being presented incorporates a channel-wise attention mechanism to dynamically assign weights to different electrodes. A Convolutional Neural Network (CNN) is used to extract the important spatial details from the encoded rEEG data. An LSTM combines extended self-attention to examine the time-frequency trends in rEEG data by evaluating significance based on built-in similarities in the signals. The proposed model demonstrated strong performance with an accuracy of 97.34% in distinguishing between depression and healthy controls using 10-fold cross-validation. According to our research, the front and occipital lobes—particularly the Fp1 and Fz locations and the O1 and O2 positions with the oscillations of beta and alpha—play the largest roles in the detection of depressive disorders.

563

Authors: Ammaiyappan Kavitha, Valavandan Srinivasan Meenakshi

Mobile Adhoc Networks (MANETs) are vulnerable to various attacks such as Black Hole Attack (BHA), Gray Hole Attack (GHA), and Wormhole Attacks (WHA). While researchers have focused on detecting and mitigating individual attacks, protection against collaborative attacks is limited. Therefore, this article introduces a new Tunicate Swarm Optimization Q-learning-based Collaborative Attacker Detection Algorithm (TSOQCADA) to identify and prevent collaborative attackers like BHA, GHA, and WHA, thereby improving routing efficiency. This algorithm utilizes feedback about node properties such as energy, reputation, buffer space, transmission delay, and packet transfer rate from all nodes to determine efficient packet routing. In the TSOQ-learning algorithm, the TSO is adopted to set the Q-table values, resulting in faster convergence of Q-learning. First, a Q-table with prior knowledge is trained to enhance searchability. Additionally, a novel selective search mechanism is adopted to improve exploration efficiency and reduce unwanted explorations by considering the correlation between current and target locations. Furthermore, a nonlinear function is designed to achieve a tradeoff between search and use abilities in Q-learning, dynamically changing ε value in the ε-greedy method according to the number of iterations. Thus, the TSOQ-learning can efficiently obtain a routing path by isolating collaborative attackers with low reputation values. Simulation results show that the TSOQCADA achieves a Packet Delivery Ratio (PDR) of 94.8%, Packet Loss Rate (PLR) of 5.2%, energy consumption of 2.53J energy/packet, throughput of 355Kbps, and End-to-End (E2E) delay of 35ms for a network of 100 nodes with 20 malicious nodes in MANET, outperforming the Efficient Trust-based Routing Scheme (ETRS), Hybrid Trust-based Reputation Mechanism (HTRM) and Deep Neural Learned Projective Pursuit Regression-based Watchdog Malicious Node Detection and Isolation (DNLPPR-WMNDI) algorithms.

586

Authors: Samar Farhan, Mohammed Aal-nouman

The deployment of small cells (SCs) in heterogeneous beyond 5G (B5G) networks holds immense promise in meeting the ever-growing demand for data rates and ensuring the desired Quality-of-Service (QoS) for cell-edge users equipment (CEUEs) in densely populated B5G networks. In our pursuit of jointly maximizing the sum-rate while minimizing interference for CEUEs, we present an optimization-based approach with a specific focus on fulfilling the minimum QoS requirements of CEUEs. Our innovative two-step algorithm begins with a reinforcement learning (RL)-based matching process among UEs and available resources, followed by optimal power allocation to UEs based on the matched pairs. At the first step, by leveraging a Q-learning-based method, our algorithm identifies the optimal UEs-resources pairing. The learning process utilizes the achieved sum-rate of each pairing among UEs and resources at each step and converges to a sub-optimal pairing among them. In the second step, power allocation for the selected pairing of the first step is solved in optimal manner. Achieving optimal power allocation is facilitated by exploiting the difference of concaves form of the objective function and harnessing the majorization-minimization (MaMi) technique considering the minimum required QoS of CEUEs.Our numerical results showcase the effectiveness of the proposed scheme, demonstrating near-optimal performance. The results show the employed RL approach effectively converges to the near-optimal pairing among UEs and resources in a dense environment. Additionally, it is evident that the optimal power allocation not only maximizes the sum-rate but also minimizes interference for CEUEs. Considering different values of macro cell (MC) transmission power and SC radius, the proposed schemes achieve a sum-rate enhancement of at least 10% and 25% compared to other existing matching and power allocation methods, respectively.

622

Authors: Ahmed Sabah Al-Araji, Khulood Eskander Dagher, Mohammed Najm Abdullah

This paper presents an adaptive neural network acting like a proportional-integral-derivative (PID) controller that uses an intelligent meta-heuristic technique to improve the drug infusion rate (propofol) as a manipulated variable in closed-loop control of anesthesia systems using the Bispectral Index (BIS) as the primary controlled variable. The effect of propofol on the human body is modelled using the pharmacokinetic (PK) and pharmacodynamics (PD) models. A physiological dataset of patients, including gender, weight, height, age, and the like, determines the parameters of the PK/PD mathematical model. The proposed controller seeks to provide the optimal propofol control action, which is in charge of swiftly, precisely, and accurately maintaining a triad of hypnosis, analgesia, and neuromuscular blockade by infusing several drugs that are specific to each state. To train this neural network like a PID controller with the radial basis function (RBF) in a neuron, the meta-heuristic method is employed. The first technique is particle swarm optimization (PSO), which has been widely used in both data estimation and training because of its quick computing speed, while the second technique is the chaotic PSO algorithm, and the third technique is the modified CPSO algorithm (MCPSO). The fundamental proposed procedures of the MCPSO algorithm use the chaos method, including the coefficients of acceleration, and remove the two random parameters from the velocity update equation to generate more randomness in the search space to quickly solve the local minima problem. The PSO, CPSO, and MCPSO meta-heuristic algorithms use the mean square error (MSE) performance index to find and optimize the optimal or the nearly ideal gain parameters of the nonlinear neural network to function like a PID controller. The simulation results show that the proposed controller for different physiological dataset patients is characterized by its efficacy and resilience in terms of controlling the depth of the hypnosis state and the infusion rate of the anesthetic drug during surgery in order to avoid under- or over-dosing of the drug for the patient through the desired value of BIS (50) with minimizing the steady-state error, which is equal to zero without any oscillation. Moreover, the comparison results showed that the proposed RBF-NN-PID controller enhanced the time in one minute to reach the depth of anesthesia at the moderate hypnotic state when compared to the fractional-order adaptive high-gain controller, in which the time to reach the depth of anesthesia is two minutes. In contrast, the adaptive neuro-fuzzy controller reached the depth of anesthesia in three minutes. Therefore, the time was improved by 50% and 67%, respectively. In particular, the surgery BIS index was kept at the BIS desired 50 at the moderate hypnotic state without any error and with no oscillation at steady-state.

647

Authors: Preeti Sadanand Topannavar, Dinkar M. Yadav, Varsha Bendre

The early analysis of brain tumors plays a significant role in enhancing treatment and patient survival rates. The accurate diagnosis is critical to creating treatment plans that extend the lifetime of affected individuals. The existing methods have limitations like less accuracy and not able to learn global features. To overcome this issue, Global Self-Attention Module based Convolutional Neural Network (GSAM based CNN) is proposed for brain tumor classification. The Figshare dataset is preprocessed through data augmentation which is utilized to enhance the data size, and the median filter is employed to eliminate noise in an input image. The preprocessed images are given to Adaptive Kernel Fuzzy C Means (AKFCM) with Otsu thresholding for segmentation process. After segmentation, the Sine Cosine Reptile Search Algorithm (SCRSA) is employed for feature selection. Then, the features are provided to GSAM based CNN for brain tumor classification. The proposed model achieves better result on Figshare dataset on the metrics of accuracy, precision, sensitivity, specificity and f1-score, with values of 99.83%, 99.65%, 99.79%, 99.78% and 99.71%, correspondingly, when compared to the existing methods like Comprehensive Learning Elephant Herding Optimization based K-Nearest Neighbor (CLEHO based KNN) and Parallel Deep CNN (PDCNN) with data augmentation.

671

Authors: Sahar K. Hassan, Zaid M. Khudair

The study presents an innovative approach to designing a Circular Microstrip Patch Antenna capable of covering a broad range of frequency bands. The physical measurements of the antenna are 34×22 × 1.57 mm³. Rogers RT5880 (Lossy) substrate has been chosen to build the antenna, which has a dielectric constant (????????) of 2.2. The antenna’s radiating patch layer and the ground plane layer has been structured from copper, each with a thikness of 0.035 mm. The resonance at five distinct frequencies: 5.2 GHz, 6.628 GHz, 9.098 GHz, 12.036 GHz, and 13.076 GHz is the key element of this antenna. These frequencies cover a broad range of applications including WLAN, C-band, X-band, and Ku-band. Simulations has been applied using the CST Software version 2021 to validate the antenna performance. The adaptability of the antenna design, coupled with its successful validation, makes it suitable for a broad range of wireless applications across different frequency bands. This highlights their effect and applicability across a variety of technological fields.

696

Authors: Harikrishna Mulam, Venkata Rambabu Chikati, Bhargav Salugu

The Electrocardiogram (ECG) serves as a crucial indicator of diverse cardiac conditions, emphasizing the importance of precise signal classification for automated arrhythmia detection. ECG is an efficient tool for the diagnosis and detection of arrhythmia. Detecting arrhythmias in extended ECG segments can result in episodes being overlooked. However, since ECG transmits a massive amount of information, it becomes very complex and challenging to extract the relevant information from visual analysis. To overcome this problem, this research proposes Long Short-Term Memory (LSTM) with Luong Attention Mechanism approach for the classification of arrhythmia from ECG into 5 classes. When LSTM is combined with Luong attention, they can learn which parts of the ECG signal are crucial at each time step, and effectively capturing both short-term and long-term dependencies. For evaluating the performance of the proposed method, the data is collected from a benchmark dataset called the MIT-BIH dataset. After the collection of the dataset, the pre-processing is done using Continuous Wavelet Transform (CWT) to reduce the low and high-frequency noise. After that, the pre-processed data is forwarded to the feature extraction process to extract the relevant features by using the statistical (Skewness, Kurtosis, Moment, etc.) and time-frequency domain features. Finally, the LSTM is used to classify the arrhythmia classes. From this analysis, the proposed LSTM with Luong Attention Mechanism achieved better results in overall metrics. The proposed method achieves the accuracy of 99.75% which is comparatively higher than the existing approaches like Deep Residual Convolutional Neural Network (DRCNN) and Depth wise Separable CNN with Focal Loss (DSC-FL-CNN).

720

Authors: Ratnam Dodda, A Suresh Babu

Text document clustering (TDC) is becoming the most famous technique in various real-world applications like information retrieval, data mining, and so on. The TDC poses a significant challenge in these applications as the volume of data collection continues to grow daily. In clustering, the metaheuristics algorithms are used for solving the NP-hard problem and to enhance the model performance. Once the solution space is maximum, traditional approaches are incapable of identifying the solution in minimal time. In this research, the Chaotic Northern Goshawk Optimization (CNGO) with K-means Clustering algorithm is proposed for TDC. Initially, three benchmark datasets such as Reuters-21578, 20-Newsgroup and BBC Sport are used to estimate the performance. Then, the pre-processing techniques such as tokenization, stop word removal, stemming and lemmatization are used for data denoising. The word2vec feature extraction technique is used for converting the document into numerical vectors and finally, the CNGO with K-means is used for TDC. The effectiveness of the proposed method is estimated by utilizing various metrics and it attains the accuracy of 98.36%, 96.26% and 97.36% by using 20-Newsgroup, Reuters-21578 and BBC datasets when compared to competitor approaches like Salp Swarm Algorithm (SSA) and Rider Optimization-based Moth Search Algorithm (Rn-MSA).

744

Authors: Ruaa Abdulkareem Yaseen, Fadhil S. Hasan

This paper introduces a new index modulation method called Grouping Frequency-Time Index Modulation Non-Active for Differential Chaos Shift Keying (GFTIM-DCSK-II). The proposed system exploits the advantages of index modulation (IM) so that both the energy and spectrum efficiency of the system are enhanced. The suggested system effectively employs both frequency and time resources, allowing for efficient transmission at large data rates. For each symbol period, the new approach divides all data bits into equal G groups. For each group, the system includes frequency index bits, denoted as m_1, which are utilized to choose a non-active subcarrier from a total of 2^(m_1 ) subcarriers. Additionally, time index bits, denoted as m_2, are employed to select a single time slot from 2^(m_2 ) total time slots. After that, the modulated bits m_0 are subjected to DCSK modulation and thereafter sent over the active subcarriers. Therefore, G(m_0+m_1+m_2), where m_0=(2^(m_1 ) 2^(m_2 )-1), is the total amount of bits transmitted with the GFTIM-DCSK-II system. To better demonstrate the distinctive characteristics of the proposed scheme the energy efficiency, complexity, and spectral efficiency of the GFTIM-DCSK-II system in comparison to the most advanced alternatives available are thoroughly analyzed. The system with q=8, m_1=1, and m_2=1 saves around 40% more energy than the CTIM-DCSK system also the proposed system enhanced the spectral efficiency around five times more than CTIM-DCSK and GSIMDCSK-II with q=8, m_1=1, and m_2=4. Bit error rate (BER) formulas across multipath Rayleigh fading channels and additive white Gaussian noise (AWGN) are calculated analytically. The proposed system showed an enhancement in BER performance around 5 dB than CI-DCSK2 and GSTIM-DCSKII across the AWGN channel at a BER of 10^(-4) at the same data rate.

771

Authors: Saad M. Alwash, Osama Qasim Jumah Al-Thahab, Shamam F. Alwash

The paper presents an algorithm for harvesting optimal power points (OPP) utilizing the direct power control (DPC) technique of the dual-fed induction generator (DFIG) wind turbine. The conventional Fibonacci search algorithm (FSA) technique cannot find the exact global peak (GP) strategy and may fail. It proposes a new modified Fibonacci search algorithm (MFSA) to extract (OPP) with variable-speed wind energy turbines (VSWET), which is very important for many people and companies. Among the other different algorithms, the MFSA is more efficient and has a simple strategy to track the optimal power point that depends on calculating the effective speed of the Ω_m in an interval [a, b]. The research paper suggests modifying the Fibonacci-based search on the rotor side conversion (RSC) of a 7.5 KW DFIG in the WECS wind energy conversion strategy. The simulation uses the built-in MATLAB/SIMULINK (R2023a) software. It demonstrates the ease of implementation, lower total harmonic distortion THD and efficiency since it uses a step size that is not fixed based on the Fibonacci sequence.

802

Authors: T. Divya, J. Viji Gripsy

This study proposes three contributions focused on lung cancer detection and severity level identification. The absence of non-invasive technologies for predicting lung cancer necessitates faster, more efficient, and more accurate classification procedures due to the absence of non-invasive technologies for predicting lung cancer. Creating an automated and intelligent prediction system is crucial for identifying phases and predicting the possibility of a recurrence. The objective is to create an automated detection system for identifying lung cancer using an optimization-focused deep learning model. We develop an adaptive multi-swarm PSO and combine it with the firefly algorithm to determine the ideal weight values for the Wavelet Neural Network (WNN) model. We use the HAPSO-FFA-WNN method to explore problems with multiple optimal solutions. This study evaluated two lung cancer datasets, and the proposed HAPSO-FFA-WNN model achieved 97.58% accuracy for dataset 1 and 98.54% accuracy for dataset 2. Furthermore, the proposed model achieved better precision, recall, and MCC performance metrics.

26

Authors: Maddu Srinivasa Rao, D Nagendra Rao

One of the key elements of the internet of things (IoT) nowadays are wireless sensor networks (WSNs). Any IoT application uses a variety of sensor-based devices to collect data from real-world objects and send it to the base station (BS). In accordance with the sensor position, the BS evaluates this data. Any IoT-based application needs a high-performance intelligent WSN. IoT-enabled WSNs are high-performance intelligent WSNs, which are used throughout this article. Fault incidence likelihood in IoT-enabled WSNs is substantially higher than in conventional networks. The dependability of the IoT-enabled WSNs is affected by faulty nodes and broken connections. Incorporating multipath transmission, relay node location, and backup node selection, various fault-tolerance techniques improve the network's dependability. These methods, however, have significant packet overhead, worse detection accuracy, and lengthy data transmission delays. For fault tolerance in IoT-enabled WSNs, a multi-objective red panda optimization method incorporating deep reinforcement learning (DRL) is suggested in this paper. This study's primary goal is to identify defective nodes using starting energy, transmission of packets rate, communication overhead, and packet delay measurements. In order to capture data in an energy-efficient manner and extend the network lifespan, a mobile sink (MS) is deployed. The suggested approach beat state-of-the-art techniques in terms of energy efficiency (EE), latency(L), packet delivery ratio (PDR) and network lifetime (NL) according to thorough simulations and theoretical research.

50

Authors: Sunil A. Patel, Pratik A. Barot

Machine learning models have been used to protect the plant from leaf disease by early detection and classification in the field of agriculture. Farmers' principal source of income is agriculture, and plant leaf disease causes them to lose a lot of crop each year. It is essential to provide food for the whole world by detecting disease at an early stage in the farm for future decision-making. However, there is currently no such kind of system in place to forecast a disease at an early stage. The primary goal of this research is to inform farmers about a novel strategy for preventing leaf disease in plants. Through this research and earlier diagnosis, the losses incurred by farmers might be decreased. In this study, we have suggested a unique custom parallel deep convolutional neural network-based model for the classification of leaf diseases on the farm. This proposed CNN model is trained using the plant village dataset with 10 different categories of leaf disease classes. The patterns of leaf images at particular times are utilized in conjunction with computer vision techniques to detect plant leaf diseases. The tomato plant is taken into consideration for current scientific efforts in disease identification, categorization, and diagnosis. Image resizing, cropping, segmentation using k-means clustering and thresholding, and normalization are five different types of data preprocessing techniques used. The suggested approach may accurately diagnose diseases by automatically extracting features using implicit CNN. During comparisons with the testing of the given data, the output of the proposed model is 99.65%, which is better than transfer learning models available in the literature. This suggested model may be checked for consistency and reliability, and farmers can use it as a practical tool to protect their leaf plants from any kind of disease.

71

Authors: Manish Shankar Kaushik, Aditya Bihar Kandali

In today's digital landscape, the threat of digital image forgery is on the rise, necessitating the development of advanced detection and authentication techniques. Copy-move image forgery, a prevalent form of manipulation, involves duplicating a portion of an image and inserting it elsewhere with malicious intent. Detecting such forgeries is of utmost importance. This research introduces an innovative approach that combines convolutional neural network (CNN)-based deep features and gray-level co-occurrence matrix (GLCM) features. These hybrid features are carefully selected using neighborhood component analysis (NCA) to optimize their discriminative power. Subsequently, a support vector machine (SVM) classifier is employed to classify these refined hybrid features, resulting in exceptional detection performance. The proposed method achieves remarkable accuracy in forgery detection. In the CASIA 1 dataset, the hybrid features-NCA-SVM method outperforms other techniques with an accuracy of 97.62%. Similarly, on the MICC-F220 dataset, the hybrid features-NCA-SVM approach attains the highest accuracy of 98.00%. These results underscore the robustness and versatility of the proposed method in detecting copy-move forgeries across different datasets.

95

Authors: Alfan Alfarisy, Handayani Tjandrasa

In recent years, radio frequency interference (RFI) has emerged as a significant challenge in weather radar systems due to the escalating demand on the RF spectrum. This interference results in distortions in radar imagery, diminishing the accuracy of weather forecasts. While current mitigation methods rely on threshold algorithms and linear detection, there's been a notable enhancement with the incorporation of deep learning techniques, especially through convolutional neural networks (CNNs). In this context, we introduce the R2U-Net3+ architecture, a fusion of U-Net3+ and recurrent residual convolutional neural networks (RRCNN). Focusing on detecting RFI-affected areas in radar images, this architecture has exhibited commendable results in tests. Based on evaluation metrics, R2U-Net3+ achieved a dice coefficient of 94.980% with a single recurrence iteration and improved to 95.352% with two iterations. For the IoU, R2U-Net3+ scored 90.745% with one iteration and 91.409% with two iterations. These results affirm that R2U-Net3+ offers a significant improvement in detecting RFI, positioning it as a leading solution to challenges faced by weather radar systems.

123

Authors: Nang Kham Htwe, Win Pa Pa

Text-to-image synthesis (T2I) is a challenging task because the model requires to create high-quality images that are semantically consistent and realistic. Therefore, the main objective of this paper is to improve the quality of generated images and the similarity level between text descriptions and these images. In this paper, we proposed deep-fusion generative adversarial networks (DF-GAN) with multimodal similarity model (MSM) to generate high-resolution images with better consistency between text and the generated images. In this work, MSM is pretrained using real images with captions in the dataset. This pretrained model is used to improve the visual-semantic consistency level during training of T2I. This paper investigates the improvement in the image generation process due to applying MSM model to the generator. The investigation is performed on two different datasets with different languages to prove our proposed model outperforms baseline DF-GAN. Firstly, the experiment is done on Caltech- Birds dataset and the evaluative results of the proposed model are compared with state-of-the-art models, StackGAN, StackGAN++, AttnGAN, DMGAN, DAE-GAN, TIME, DF-GAN. According to the comparative results, the proposed model outperforms the baseline state-of-the-art models in terms of Fréchet inception distance (FID) score and inception score. The improvements of the proposed model on synthesizing images over the baseline models are proved in terms of image quality and visual-semantic similarity in this work. Accordingly, the proposed model is applied on Myanmar text-to-image synthesis (Myanmar T2I) with Oxford 102 flowers dataset annotated in Myanmar to prove the effectiveness of the proposed model on different dataset and different language. To the extent of our knowledge, this is the first attempt of implementing generative adversarial networks with multimodal in Myanmar T2I and it got inception score of 3.54 ± 0.03 and FID score of 49.97. In Myanmar T2I, the proposed model also got better performance than the baseline DF-GAN because it achieved higher inception score and lower FID score than the baseline model. Drawing upon the findings derived from the outcomes of the experiments, it is evident that the proposed model demonstrates an enhancement in the quality of generated images within the context of the image generation.

145

Authors: Mahmoud Zaki Iskandarani

This work investigates through simulation effect of waiting time and number of slots on the effectiveness of using Slotted ALOHA in vehicular connectivity as a random access technique. The objective is to reach optimum level of throughput under the vehicular dynamic environment. The presented work proposes to concentrate on waiting time as a controlling parameter that can be used in an adaptive manner to enable throughput optimization in response to changing vehicular density. The work proved through MATLAB simulation the importance of waiting time parameter in controlling number of delivered frames and throughput. The results show that based on the exponential relationship between throughput and data traffic, there is a logarithmic relationship between data traffic and waiting time, which complements another logarithmic relationship between throughput and waiting time. The simulation also shows that there is an inverse relationship between throughput and data traffic as a function of set waiting time. The simulation shows that there is an exponential relationship between waiting time and maximum number of delivered frames. The work shows that a linear region of operating Slotted ALOHA can be specified, such that an optimum and dynamically adjusted waiting time is set with overall effect on number of slots and throughput. The dynamic shape function of the curve correlating waiting time to available number of slots is mathematically modelled and differentially controlled, thus an optimizing and controlling algorithm is produced, which improves significantly the conventional ALOHA performance and has an advantage over other used ALOHA derivatives, in that it concentrates on available slots and waiting time, which greatly contributes towards reduction in collisions and improving of throughput.

171

Authors: Quyen Thi Nguyen

This paper shows a novel method relied on growth optimization (GO) algorithm for searching the shortest tour length of the travelling salesman problem (TSP). GO is a recent algorithm relied on the idea of learning and reflecting of people in the society. To enhance performance of GO, the 2-opt local search technique is applied for adjusting the candidate solutions created by GO. The effectiveness of GO is validated on five TSP instances consisting of the 14-city, 30-city, 48-city, 52-city and 76-city. The error between the optimal tour length value obtained by GO and the best-known value for these instances is 0.0000%, 0.0000%, 0.0021%, 0.0314% and 0.0004%, respectively. Furthermore, the comparisons to the methods in the literature in term of the optimal and mean tour length values have shown that GO reaches the better values compared to other prvious methods. Thus, the proposed GO approach is a potential method for the TSP problem.

190

Authors: Miftahul Hasanah, Rizqy Ahsana Putri, Muhammad Aidiel Rachman Putra, Tohari Ahmad

The evolution of technology and the internet has accelerated the pace of communication and information exchange. Despite technological advancements, the significant weakness lies in the persistent threat of cybercrime, manifesting in various forms like malware, phishing, and ransomware. To solve the cybercrime problems, this research aims to create an intrusion detection system model using a novel framework. In general, the proposed method consists of 3 stages: Data preprocessing, feature selection using ANOVA F-value combined with cross validation, and classification using weight-based voting classifier. Some machine learning methods used in the weight-based voting classifier are random forest, K-nearest neighbour, and logistic regression. The experiment results show that weight order and weight combination affect the detection performance. The proposed method produces an excellent precision value of 98.66%, higher than the single voting classifier.

210

Authors: Arna Fariza, Rengga Asmara, Eha Renwi Astuti, Ramadhan Hardani Putra

Abnormalities commonly encountered in dental practice include tooth and supporting tissue issues such as caries, periapical abnormalities, resorption, and impacted third molars. Panoramic radiographs are frequently used for image scanning in dentistry and oral surgery. Diagnosing dental anomalies can be time-consuming due to the complexity of the orthodontic area, potentially leading to inaccuracies. This research proposes an end-to-end automated detection of dental and supporting tissue anomalies in patients, encompassing cavities, periapical lesions, resorption, and impacted third molars. This study evaluated the effectiveness of employing various pre-trained Convolutional Neural Network architectures, including ResNet-50, ResNeXt-50 32×4d, Inception-V3, and EfficientNet-V2. To enhance model performance, a batch normalization technique was integrated into the classification layer of these pre-trained models. Data pre-processing techniques, including horizontal and vertical flips, as well as random affine transformations, were applied to augment the dataset. Additionally, an image normalization procedure was implemented before the training and prediction phases. In the evaluation on 202 images, the integrated ResNeXt-50 32x4d model with batch normalization achieved the highest accuracy, precision, recall, and F1-score of 83.663%, 81.615%, 81.271%, and 81.066%, respectively. Based on the F1-score, this model demonstrates promising predictions of tooth and supporting tissue anomalies in an imbalanced dataset.

235

Authors: Made Liandana, Dandy Pramana Hostiadi, Gede Angga Pradipta

Human Activity Recognition (HAR) is focused on Activities of Daily Living and developed in the health and human security fields. The HAR concept was introduced in previous research using multi-sensor devices. In their implementation, wearable devices require computational and real-time environmental limitations. This paper proposed a new approach for HAR using a machine learning-based single-sensor accelerometer. This research aimed to determine the performance of machine learning in HAR using three Feature Subsets: Feature Subset Signal Vector Magnitude (SMA), Feature Subset Fast Fourier Transform (FFT), and Feature Subset Value-Crossing. In features selection, ANOVA was used to reduce feature dimensionality. The experimental results have been assessed using the confusion matrix to prove that the proposed model can achieve an optimal accuracy of 0.97, higher than several state-of-the-art approaches. The optimal sensitivity and specificity values have been 0.98 and 0.99 and are partially higher than previous studies using similar testing scenarios.

266

Authors: Nagajyothi Dimmita, Vanga Nagasri, Koppaka Achutha Jyotsna, Pochaboina Swapna, Narne Srikanth, Pattem Sampath Kumar, Atheeswaran Athiraja, Gunaganti Sravanthi, Rajeswaran Nagalingam

Breast cancer is the most typical cancer overall and the one that affects women the most frequently. In 2022, there were 2.26 million or more brand-new cases of breast cancer in women. To identify the breast cancer in early stage, it’s easy to cure. The apparent structure is examined using computer-aided diagnostic (CAD) technologies. The majority of existing diagnostic approaches rely on mammography mass features; however, the suggested strategy relies on MicroCalcification (MC) characteristics. In this research cancer cells are detected utilizing mammography images, pre-processing methods, segmentation, multi-layer perceptron with artificial algae algorithm and receiver operating characteristics (ROC) curve analysis. The national cancer institute (NCI) of provides mammogram scans. Normalization and median filtering were used as part of the pre-processing procedure. Segmentation is used to detect the location of MicroCalcification present cells using the local threshold approach. The MicroCalcification present cells and MicroCalcification missing cells are classified using the multi-layer perceptron (MPL) with artificial algae algorithm technique and also used to divide MicroCalcification-affected cells into the following categories: initial, very small, small, medium, high, and very high. ROC curve analysis was used to assess system performance. According to experimental findings, for the NCI, University of California Irvine (UCI), nathan kline institute (NKI), investigation of serial studies to predict your 1 (ISPY1), ISPY2, ISPY3, and ISPY4 datasets, the classification using the multi-layer perceptron with artificial algae method has the best accuracy of 96 percent when compared to random forest (RF), MPL with genetic algorithm (GA), hierarchical clustering random forest (HCRF), and convolutional neural network (CNN).

294

Authors: Hanaa M. Al Abboodi, Amera W. Al-funjan, Ahmed Aldhahab, Wafaa M. Salih Abedi, Alaa H. Abdullah, Ameer Hamza Siraj

The research presents a new technique for segmenting brain tumors using the UNet framework enhanced with an attention mechanism. By incorporating attention processes that selectively emphasize prominent aspects while recording comprehensive contextual information, our strategy overcomes the challenges of brain tumor delineation. The suggested UNet-attention model is intended to outperform traditional segmentation techniques regarding precision and clinical applicability. Integrating spatial and channel attention processes into the UNet design is one of our study's significant achievements. The spatial attention mechanism's focus improves the capacity of the model to differentiate the mechanism's tumor and non-tumor areas. Also, incorporating contextual clues from multi-scale hierarchies allows for a thorough comprehension of visual properties. The discrete wavelet transform has been applied as a feature extraction method to enhance the model performance regarding time and memory consumption. A wide range of datasets is evaluated in-depth, proving our UNet-attention model's superiority. Advanced deep learning is made possible by combining attention processes and contextual data to delineate tumors precisely and clinically. Many evaluation criteria involving dice scores, accuracy, mean IoU, sensitivity, specificity, and Hausdorff distance have been applied to evaluate our model performance in different aspects. The model attained a dice coefficient of 0.9971. The model's specificity of 0.9988 is particularly noteworthy, demonstrating its exceptional ability to identify regions without tumors accurately. The model also achieved 0.9986 accuracies, 0.9142 mean IoU, Hausdorff distance (mm) 3.48. These evaluation values were obtained for applying our model on flair images from BraTS 2020.

324

Authors: Arumugam Gayathiri, Sabhapathy Mohanapriya

The mobile adhoc network (MANET) has attracted considerable attention from researchers due to its dynamic and versatile nature. Graph clustering algorithms can be more effective than optimization algorithms in modeling and analyzing networks because these algorithms arrange nodes into clusters based on their connectivity. According to this context, a graph kernel-based clustering algorithm (GKCA) was developed for MANETs by combining the d-hop graph kernel and clustering scheme. Additionally, it uses the shortest route to connect multiple cluster head (CH) nodes for data transfer. MANETs face challenges such as changes in network structure and disruptions in communication links, which result in an increase in route discovery requests and longer mean end-to-end delays (MED) due to longer link reconnect times. Hence, this article proposes the GKCA with link failure prediction (GKCA-LFP) in MANETs to prevent path failures resulting from node mobility. The GKCA is initially used to determine the cluster size and CH nodes. The shortest route is used to connect the CHs for data transfer. Then, the LFP strategy is introduced at this stage to maintain the path. This strategy aims to predict the current link status based on mobility and position information to prevent failure conditions and minimize packet loss ratio (PLR). The GKCA-LFP algorithm can choose more stable shortest paths to connect CHs for data transfer, resulting in decreased MED and PLR. The extensive simulations show that the GKCA-LFP algorithm outperforms the GKCA, AMAC, MARP-HO, and RS-GG algorithms in MANETs. Specifically, for 100 nodes, the GKCA-LFP algorithm achieves a 1.4% control packet ratio (CPR), 0.8% PLR, and 455µs MED. Additionally, for nodes with a mobility speed of 20m/s, the GKCA-LFP algorithm achieves a 1.2% CPR, 2.3% PLR, and 120µs MED.

346

Authors: Sura Muhi Hussein, Ahmed Sabah Al-Araji

As a result of the widespread adoption of hovercraft systems, the path finding of such systems has become an essential rule for locating the target with the shortest distance and avoiding collision between the starting point and the target locations. The purpose of this paper is to develop a method for path finding by proposing a hybrid method of intelligent optimization techniques, including two stochastic approaches. The first one is called the Artificial Bee Colony (ABC) algorithm, and it aims to direct the hovercraft toward the target utilizing the behaviour of honey bees to reduce the path length required to reach the target. The second approach makes use of the Self Perception Particle Swarm Optimization (SP-PSO) algorithm, which is designed to improve particle swarm optimization in order to obtain a path that is more effective, smoother, and shorter for the hovercraft to travel in a complicated environment. The developed hybrid method is called the Artificial Bee Colony Self Perception Particle Swarm Optimization (ABC-SPPSO), which enhances the convergence speed and achieves a trade-off between exploration and exploitation in order to generate the shortest path while simultaneously avoiding collisions. The simulation results indicate that the performance of the proposed hybrid algorithm surpasses those of the original techniques. Moreover, the results demonstrate that the proposed hybrid algorithm outperforms the hybrid Firefly Algorithm Modified Chaotic Particle Swarm (HFAMCPSO) by (11.90%) in terms of distance and (56%) in terms of iterations. In addition, it outperforms the Quarter Orbit combined with the Particle Swarm Optimization (QOPSO) algorithm by (1.89%) in terms of distance and (16.66%) in terms of iterations. Furthermore, compared to the Rapid Random Tree Star Particle Swarm Optimization (RRT*PSO), the proposed method achieved enhancement of (0.82%) in terms of distance and (33.33%) in terms of iterations in providing the shortest path while avoiding collisions and reducing the trajectory tracking error to approximately zero. All the considered approaches were simulated in a global environment, utilizing the MATLAB 2022 package.

377

Authors: Purba Daru Kusuma, Ashri Dinimaharawati

This paper presents a new metaphor-free metaheuristic search called the swarm bipolar algorithm (SBA). SBA is developed mainly based on the non-free-lunch (NFL) doctrine, which mentions the non-existence of any general optimizer appropriate to answer all varieties of problems. The construction of SBA is based on splitting the swarm into two equal-sized swarms to diversify the searching process while performing intensification within the sub-swarms. There are two types of finest swarm members: the finest swarm member for the whole swarm and the finest swarm member in every sub-swarm. There are four directed searches performed in every iteration: (1) search toward the finest swarm member, (2) search toward the finest sub-swarm member, (3) search toward the middle between two finest sub-swarm members, and (4) search relative to the randomly picked swarm member from another sub-swarm. The performance of SBA is assessed through two assessments with a set of 23 functions representing the optimization problem. In the benchmark assessment, SBA is contended with five metaheuristics: northern goshawk optimization (NGO), language education optimization (LEO), coati optimization algorithm (COA), fully informed search algorithm (FISA), and total interaction algorithm (TIA). The result presents the superiority of SBA among its contenders by being better than NGO, LEO, COA, FISA, and TIA in 21, 16, 16, 21, and 18 functions. The single search assessment is performed to evaluate each strategy involved in SBA. The result shows that the search toward the middle between the two finest sub-swarm members is the best among the four searches in SBA.

402

Authors: M. V. Ramesh, K. Swarnasri, P. Muthukumar, Ponnam Venkata Kishore Babu

The paradigm shift towards sustainable energy sources has propelled research into enhancing the efficiency and reliability of power systems. This study introduces an advanced methodology for optimal power flow (OPF) by integrating the skill optimization algorithm (SOA) with considerations for open-access trading of wind farms and the integration of electric vehicle (EV) fleets. In order to improve exploration features of SOA, opposition-based learning (OBL) is used for diversifying the population. The proposed ISOA-based OPF model addresses the optimization of power flow in a multi-objective framework, aiming to minimize generation cost and minimize transmission loss while accommodating the open access trading between wind farms and EV fleet demands. The incorporation of open-access trading enables the effective utilization of surplus wind energy among interconnected power systems, fostering grid resilience and sustainable benefits. Simulation results on standard IEEE 30-bus and 57-bus test systems validate the efficacy of the proposed approach, showcasing improved system performance, reduced cost and power losses, and enhanced utilization of renewable resources in modern power grids. In the IEEE 30-bus, fuel costs are $803.13/hr standard and $935.2408/hr with extra trading. In the IEEE 57-bus, costs are $37589.34/hr standard and $37628.8/hr with additional trading. These results align with benchmarks, showcasing the method's efficacy in intricate problem-solving.

429

Authors: Lalitha Kondisetti, Katragadda Swarnasri

In today's electrical distribution systems (EDS), the rise of renewable energy sources (RESs) and electric vehicles (EVs) holds immense significance. Despite their positive environmental impact, these technologies pose challenges due to their unpredictable nature. Efficiency and reliability are top priorities for EDSs, which can be achieved through optimal network reconfiguration (ONR). This approach addresses integrating varying RESs and EVs levels by considering distribution losses and the line loadability index (LLI). To enhance network performance by identifying optimal branches and tie-lines for switching, a novel meta-heuristic algorithm, the northern goshawk optimization algorithm (NGO), is proposed. To boost its search capabilities, an improved version, called improved NGO (INGO), introduces a levy flight distribution and an adaptive parameter. Simulations on unified Egyptian network (UEN) system across diverse scenarios demonstrate INGO's computational efficiency surpassing basic NGO and other algorithms like stochastic fractal search (SFS), harmony search algorithm (HAS), and artificial rabbits algorithm (ARO). INGO outperforms in target function optimization and computation time, showcasing its adaptability for real-time applications through reduced losses and enhanced loadability. In the initial network configuration, real power losses were measured at 805.73 kW and reactive power losses at 361.18 kVAr. Notably, the lowest voltage magnitude was recorded at bus-30, reaching 0.9463 p.u. The reliability index SAIDI was calculated as 3.148, corresponding to a maximum loadability of 1.73101 p.u. Following optimal reconfiguration with INGO, significant enhancements were observed. Losses reduced notably to 768.37 kW for real power and 360.86 kVAr for reactive power. The lowest voltage at bus-25 increased to 0.9511 p.u. Furthermore, SAIDI improved to 3.001, while loadability increased to 1.73903 p.u. These improvements were consistent even with varying PV and EV load penetrations.

453

Authors: Thakur Saikumari, George Victo Sudha

Nowadays, blockchain-based healthcare systems have gained more attention for handling a huge amount of patient health data observed by Internet-of-Things (IoT) systems. Several studies have focused on integrating blockchain technology with healthcare systems to ensure the secrecy and transparency of patient data. Among these, an enhanced eHealthChain system has been designed, which comprises Enhanced OAuth (EOAuth) 2.0 to enhance user security based on their trust score. It also applies a Constrained Application Protocol (CoAP) to achieve reliable transmission by controlling congestion in the network according to the Retransmission Timeout (RTO). However, the CoAP causes long inactive latency in the network, and the backoff values are constant, which do not differ in dynamic network settings. Hence, this article proposes an Enhanced CoAP (ECoAP), which adopts adaptive RTO estimation with an Adaptive Backoff Factor (ABF), Data Loss Rate (DLR) estimation, and a Refined Random Early Identification (RREI) scheme for Congestion Control (CC) in dynamic network configurations. This can be useful to reduce the number of retransmissions by predicting the variable backoff values. Moreover, this protocol achieves effective queue management and regulation of backoff values based on congestion levels. Finally, the implementation findings exhibit that the ECoAP outperforms the classical CoAP for secure and reliable transmission in blockchain-enabled clinical IoT systems. The results revealed that the proposed system achieves 24.6% unsuccessful transmissions, 4.64 seconds average service time, 0.2264 seconds average network delay, 94.96% average server usage, 16.9% average Quality of Experience (QoE), 92.4% confidentiality, 86.5% authorization, and 93.9% data integrity for 2500 transmissions.

477

Authors: Anggun Fitrian Isnawati, Wahyu Pamungkas, M. Panji Kusuma Praja

The high mobility associated with high-speed train (HST) results in nonstationary channel conditions that characterized by varying environments leads to challenges like rapid changes in signal strength and quality. As train speeds increase, the Doppler shift occurs and the Peak to Average Power Ratio (PAPR) in the system also increases, which is a common challenge in modulation schemes such as Orthogonal Frequency Division Multiplexing (OFDM). The paper investigates the impact of HST environments on the performance of Digital Video Broadcasting - Terrestrial (DVB-Terrestrial) systems. It primarily focuses on OFDM and evaluates various performance metrics like Bit Error Rate (BER), Complementary Cumulative Distribution Function (CCDF), PAPR, and the 64-QAM constellation diagram in different HST speeds. The study includes a comprehensive analysis of the HST channel and its integration with the DVB-Terrestrial system. Key findings demonstrate how higher speeds result in increased PAPR values and changes in the 64-QAM constellation diagram, indicating the necessity for robust power amplifiers and efficient symbol detection methods. The paper also shows that while higher HST speeds generally lead to worse BER values, the performance of channel coding can mitigate this effect to some extent. The research provides significant insights for optimizing DVB-Terrestrial communication systems in HST environments, highlighting the importance of channel coding and modulation in maintaining signal quality and system efficiency.

501

Authors: Sangapillai Banupriya, Palaniappan Sharmila

In healthcare systems, blockchain technology plays a crucial role in transmitting COVID-19 data among multiple entities. Over time, various blockchain-based medical applications have emerged to handle medical information confidentially. One such system is the Scalable eHealthChain system (SeHealthChain), which utilizes a sharding scheme consisting of transaction chain and reputation chain structures to enhance throughput and security. However, the system employs a modified Raft-based Synchronous Consensus Scheme (RSCS) for generating the transaction blockchain, which can potentially introduce illegitimate transactions to the Hyperledger fabric network if a rogue node transfers them to the orderer. This poses a significant security risk in the worst-case scenarios. Additionally, as the hash rate fluctuates exponentially, the generation period of transaction blocks and computation difficulty increase. To address these issues, this article proposes an Optimized SeHealthChain (OSeHealthChain) system. It integrates a Tuna Swarm Optimization Algorithm (TSOA) with the modified RSCS to dynamically adjust the blockchain parameters in response to significant changes in the hash rate. The TSOA optimizes two variables, namely the Block Interval (BI) and Difficulty Adjustment Interval (DAI) of the Proof-of-Work (PoW) for the transaction blockchain, based on objective functions that consider the Standard Deviations (SD) of the mean BI and difficulty. By selecting appropriate variables, the system generates new transaction blocks with minimal nodes and overhead, effectively validating transactions and blocks to enhance the security level. Extensive simulations show that the OSeHealthChain achieves a throughput of 3918tps and a user-perceived latency of 63.8s for 1000 nodes, outperforming the SeHealthChain, eHealthChain, Permissionless Proof-of-Reputation-X (PL-PoRX), and hybrid Proof of Stake-Practical Byzantine Fault Tolerance (POS-PBFT) algorithms in blockchain systems. It also achieves throughputs of 7051tps, 6418tps, and 6290tps for simple, camouflage, and observe-act attacks, respectively, with 1000 nodes and a shard dimension of 200 during 20 epochs.

526

Authors: Vanitha S, Jayashree R

The higher education system (HES) in every country depicts the progress and prominence of that nation. So, the government endures the most care at all levels to enrich the quality of education on both the educator and learner side. The new academic policies also urge for worthwhile education. The HES abides by multiple steps for students' welfare in the academic curriculum, including a revamped syllabus, teaching methodology, and evaluation system. Nevertheless, students' performance is falling yearly, particularly in undergraduate programs. After COVID-19, students' study behaviour transformed abruptly due to online classes where students use mobile phones for education. They spend valuable time on the internet and gaming applications, which makes the students addicted. The evolution of Artificial Intelligence and massive student data permits us to get back the next generation by doing periodic assessments during their study period. This paper proposed a blended deep learning binary classification model (Ed-Net) using convolutional neural network and bidirectional long short term memory to predict the students' performance. Multivariate time series (MTS) student academic data is employed to train the model. To accomplish this systematic research, we followed two stages in the experiment. Stage one identifies the superlative student input data, approach (tabular/time-series), and algorithm (machine/deep learning) for better classification. Stage two executes the proposed system (Ed-Net) to attain the highest accuracy with less classification error. The synthetic minority oversampling technique (SMOTE) is applied to balance the students' pass-fail ratio. Finally, the experimental result exhibits the proposed method surpassed the baseline models with 98% accuracy, 97% precision, 94% recall, and 95% fl-score. The proposed model also uses a benchmark dataset to simulate the data and evaluate its efficacy. Moreover, any educational institution can quickly fit the academic data into this model to identify the students lacking in studies early for giving proper intervention to the parents, teachers, and students.

561

Authors: Chockalingam Arumugam, Krishnan Nallaperumal, Saravanan Arumugam

The field of aspect-level sentiment classification has gained significant attention in recent years, but limited image datasets have hindered the effectiveness of neural network models. To address the issue, we present a novel approach, Feature Pyramid Network Gated Channel Transformation (FPN_GCT) that can automatically extract highly fine-grained sentiment information from images. Our approach utilizes Gated Channel Transformation (GCT) for accurate image sentiment analysis and incorporates ResNet and Reduced Layer to enhance the model's novelty. We validate the performance of our model on the Twitter and CrowdFlower sentiment analysis datasets and demonstrate that it outperforms existing techniques including VGG-19, DenseNet121, ResNet50V2, SVM, MemNet, and RAM, with an accuracy of 91.21 for the CrowdFlower dataset and 74.47 for the Twitter dataset. Our approach offers a substantial advancement in the field of image sentiment analysis and has the potential to enhance the accuracy of aspect-level sentiment classification tasks, paving the way for more accurate and efficient sentiment analysis in various applications.

585

Authors: Amith Shekhar Chandrashekhar, Niranjan Murthy Chandrashekarappa, Puneetha Bandalli Hanumanthagowda, Anupkumar Manohara Bongale

An efficient task scheduling plays an important role in facilitating the virtual resource in a cloud computing environment by minimalizing make span and enhancing the allocation of resources. Requests for resources are treated as tasks, and appropriate resources are allocated based on user requirements. But, due to high demand and requests, the cloud has difficulty allocating resources. To overcome the issues, this research introduced an optimization-based task scheduling approach. The Multi-Objective Prairie Dog Optimization (MOPDO) algorithm is introduced which considers the makespan time and the execution time as the major objective while allocating resources in IoT. The proposed MOPDOA effectively allocates the resource to the Virtual Machines (VMs) by choosing the host with maximized resources. The search mechanism with the help of MOPDO helps to detect a suitable VM for resource allocation will be continued. After the process of allotting the resources to VMs, the load balancing process must be initiated to schedule the tasks for VMs. When the task count is assigned as 100, the makespan time of MOPDOA is 12s while Particle Swarm Gray Wolf Optimization (PSGWO) obtains a makespan time of the 20s. Similarly, for different VMs, the proposed approach is 175.45s for execution whereas the existing Improved Multi-Objective Multi-Verse Optimizer obtains 186.33s to execute for 10 VMs.

609

Authors: Dana Marsetiya Utama, Bagus Dzulhi Sanjaya, Adhi Nugraha

In the context of growing global energy demand, the industrial sector has become one of the significant contributors to the world's energy consumption. To face this challenge, scheduling has been identified as one of the potential methods to reduce energy consumption in industrial operations. This article introduces the mountain gazelle optimizer (MGO) algorithm as a solution to solve the no-wait flow shop scheduling problem with a focus on the main objective of minimizing energy consumption. The research involves comparing the performance of the MGO algorithm with popular algorithms such as grey wolf optimizer (GWO), particle swarm optimization (PSO), and genetic algorithm (GA). In addition, this research also compares the proposed MGO algorithm with the latest advanced algorithms, such as coati optimization algorithm (COA) and fire hawk optimizer (FHO). In this work, the six algorithms were tested on three different scheduling cases by repeating the process 30 times, using a population of 200 and 200 iterations to minimise energy consumption. The performance comparison between these algorithms was analyzed using the One-Way ANOVA statistical test. Based on the results, MGO outperforms GWO, PSO, GA, COA, and FHO algorithms in solving scheduling problems, with the primary function of minimizing energy consumption in Cases 1, 2, and 3. In addition, based on the convergence curve, the MGO algorithm has a better convergence curve compared to GWO, PSO, GA, COA, and FHO; it shows that the intestinal algorithm can reach the optimal solution faster and more stable during iterations than the GWO, PSO, GA, COA, and FHO algorithms. This finding confirms that the MGO algorithm has the potential to be an effective alternative in the scheduling optimization process, significantly reducing energy consumption in the industrial sector.

632

Authors: P. Kavitha, D. Dhinakaran, G. Prabaharan, M. D. Manigandan

In the rapidly evolving landscape of medical imaging, our proposed work presents an innovative and efficient approach to brain tumor detection through advanced deep learning methodologies. Central to our methodology is the strategic utilization of pre-trained weights from the formidable MBConv-Finetuned-B0 model, initially honed on the expansive ImageNet dataset, providing a foundation rich in general visual knowledge. Our subsequent fine-tuning process targets specific layers relevant to brain tumor detection, introducing two distinct convolutional layers, MBConv 6, 55, and MBConv 6, 30, meticulously added to the MBConv-Finetuned-B0 base model. These layers are intricately designed to extract and refine features specific to brain tumors, ensuring a nuanced understanding of pathology and enhancing the model's discrimination and accuracy. The flexibility of our methodology is exemplified by the thoughtful consideration of two fine-tuning options: one that adjusts all layers of the model and another that selectively fine-tunes only the proposed layers. We conduct a detailed comparative analysis, including homogeneity and median feature values, placing our work in direct comparison with established techniques such as Ensemble Transfer Learning and Quantum Variational Classifier (ETL & QVC), Ultra-Light Deep Learning (ULDL) Model, Deep Convolutional Neural Network (DCNN), and Deep Learning and Image Processing (DLIP). The results showcase the model's proficiency, achieving an accuracy of 94%, precision of 84%, recall of 92%, F1 score of 88%, and an AUC-ROC of 96%. Notably, our model demonstrates superior performance in terms of homogeneity (vE Homogeneity: 0.93, vN Homogeneity: 0.91, Enhancement Homogeneity: 0.97) and median feature values (Median vE Feature Value: 0.82, Median vN Feature Value: 0.87, Median Enhancement Feature Value: 0.80), providing a comprehensive understanding of its effectiveness in capturing subtle nuances in brain tumor images.

656

Authors: Oday Hameed Abdel Mohsen, Mohammed Kdair Abd

The most critical strategy to keep the electrical grid from collapsing is load shedding. However, by itself, this approach is unable to achieve system stabilization, as well as frequency and voltage have an impact on the network's stability as well. Traditional load-shedding designs do not take into account the various load models and the declining economic cost associated with load disconnecting. When calculating load flow, multiple load models like constant power (C.P), constant current (C.I), constant impedance (C.Z), and combined load (ZIP) may be involved, but all loads are typically described as constant electrical power (C.P). In this article, the idea of a dual approach to load shedding and capacitor placement is adopted to maximize the amount of power loss reduction in the distribution network, taking into account different load models and determining the best one that leads to the least losses, the best voltage profile, and the lowest cost. To minimize load shedding for the network and choose the optimal capacitor size and location, the Teaching Learning Based Optimization (TLBO) algorithm. The outcomes showed that using the ZIP scenario is the superior model for all scenarios in terms of reducing active power losses by about 66.018% and lowering the percentage of load shedding to 19.5195%.

683

Authors: Nishant N. Pachpor, B. Suresh Kumar, Prakash S. Prasad, Salim G. Shaikh

In current studies, India is a farming country, and the accomplishment or disappointment of the crop mainly depends on the country's rainfall design. Generally, India's farming production is primarily based on the nature of the precipitation of the rainy season rainfall. The rainy season is the primary source of water in India. Regular rainfall forecasting is the primary source for crop development. Several analyses have defined the direct effect of rainwater on harvests. The main motive of this research work is proper and early rainfall prediction, which is helpful to people who live in northeast regions inclined to natural disasters like floods, etc. It helps agriculture with decision-making in their crop and water management (WM) using extensive dataset analysis that generates maximum terms of production for farmers and profits. This proposed work introduced an improved rainfall forecasting framework, a hybrid feature selection gradient-based RNN (HFSGRNN) model with an RNN algorithm. The research uses the HFSGRNN model steps, such as initial data preprocessing steps, which are used for forecasting rainfall, handling missing value outliers, and typecasting the rainfall dataset collected from the government site. After that, an HFSGRNN method is implemented to select the valuable using stochastic gradient descent (SGD) and optimal solutions calculated by particle swarm optimization (PSO) from the preprocessed data. The hybrid optimized feature sets are fed to the rainfall forecasting of the RNN classifier. Lastly, the valuable feature sets are forecasted using decision-making, and the simulation outcome shows that the research approach performed better in rainfall forecasting. The simulation results define that the HFSGRNN model delivered the minimum value of Root means square error (RMSE= 0.10) and maximum value of accuracy rate (acc = 98.1%) compared with existing methods, such as logistic regression (LR), Long Short Term –Convolutional Neural Network (LSTM-CNN), etc. The outcomes of the research analysis will help the farmers accept efficient modeling methods for forecasting long-term seasonal rainfall.

715

Authors: Nugra Tasik Allo, Indrabayu, Zahir Zainuddin

Bounding box regression is a commonly used technique aimed at enhancing the precision of object localization, which is crucial in the field of object recognition. The intersection over union (IoU) metric, which calculates the overlap between predicted and ground truth bounding boxes, is frequently used to evaluate the performance of object detection models. However, the MSE loss function used previously is not compatible with the IoU-based evaluation and has shown sensitivity to differences in object scales. The use of IoU as the basis for loss functions has become more common in recent years, and as a result, new techniques such as the Generalized IoU (GIoU) and complete IoU (CIoU) losses have grown to be developed. This paper introduces a hybrid mechanism called GCIoU loss, which combines GIoU and CIoU losses with the aim of further enhancing localization accuracy and convergence speed. According to our findings, the average precision (AP) is greatly improved by the GCIoU loss in comparison to the GIoU and CIoU losses by 7.72% (highest) to the basis of IoU loss, and 0.87% improvement to the CIoU loss. The GCIoU loss performs consistently well across different thresholds, especially at higher levels, and has improved AP75 by 3.07% compared to the CIoU loss as the default configuration. Additionally, GCIoU loss converges faster and even more robustly in the simulation experiments by taking 14% fewer epochs than the CIoU loss, leading to localization more precisely. By this, GCIoU loss is showing its usefulness in object identification and model optimization.

10

Authors: Poornima E. Gundgurti, Shrinivasrao B. Kulkarni

The identification of users through their fingerprints has been widely used in various applications such as biometrics, border control identification, payment gateways and so on. But a proper fingerprint validation system remains a major issue in precise detection and takes more time to process the fingerprints. To overcome these fore mentioned issues, this manuscript introduced an effective fingerprint validation system using the data collected from FVC 2000 DB1, FVC 2002 DB1 and FVC 2004 datasets. The pre-processing is supported for image enhancement using CLAHE algorithm equalization and normalization. Additionally, the signal images are effectively extracted by applying the A-KAZE and SURF extracted algorithm which helps to reduce the time required for feature extraction and description of the identification process. At the final stage, random sample consensus (RANSAC) algorithm is applied to identify and validate the fingerprint effectively with better accuracy. The elapse time of the proposed approach is 3.09 s whereas the elapse time of the automatic fingerprint-based authentication framework is 4.03 s and minutiae triangulation technique is 3.55 s. Similarly, when the proposed approach is evaluated with FVC 2004 dataset, the elapse time is 5.20 ms whereas the existing scale-invariant feature transform (SIFT) detector obtained elapse time of 4.35 s.

31

Authors: Phu Ha Trieu, Thang Trung Nguyen

This study maximizes the total electric sale profit of a hybrid power system with one thermal power plant (TPP), one wind power plant (WPP), one solar power plant (SPP), and one pumped storage hydro plant (PSP) scheduled in one day. There are no inflows to the PSP, and the PSP only uses the pumped water to produce electricity. Two study cases are simulated using artificial hummingbird algorithm (AHA), war strategy optimization algorithm (WSO), and average and subtract-based optimization algorithm (ASBO). As a result, ASBO reaches the highest profit for the two cases. In addition, ASBO is also run for two other cases neglecting the operation of the PSP. The results indicate that PSP can support the system to reach a greater profit by $1859.8 and $1,577.7, respectively. Thus, the PSP is very useful for the hybrid system in reaching the maximum total profit of the electricity sale.

54

Authors: Hutashan V. Bhagat, G. Uday Kiran, Manminder Singh

Data clustering, an unsupervised machine learning technique, plays a critical part in the process of drug discovery in chemoinformatics. Researchers have come up with numerous clustering algorithms over the past decades that are well suited to analyze large chemical datasets of high dimensionality. The applications of clustering algorithms can be seen in lead compound selection which is the process of identifying the chemical compound that helps in the treatment of disease and results in the development of a new drug in the drug discovery process. The quantitative structure-property relationship (QSPR) in the drug discovery process identifies the compounds having similar properties using clustering algorithms over the structural descriptors of the chemical compounds. The quantitative structure-activity relationship (QSAR) process uses cluster analysis to identify the empirical relationships between the chemical structure and biological activities among similar compounds. The acute toxicity of the chemical compound is controlled by the chemists in the drug discovery process using cluster analysis. Considering the numerous applications of data clustering in the drug discovery process, in this paper, an improved clustering algorithm ImpClust is proposed to cluster similar compounds based on chemical composition. Five benchmark datasets are considered to evaluate the performance of the proposed ImpClust algorithm. The experimental results obtained are compared with the five commonly used clustering algorithms. A total of five cluster validation indexes (DI-Index, COP-Index, DB-Index, CH-Index and Silhouette Index) are used to evaluate the clusters formed utilizing the different clustering algorithms. The experimental findings show that the proposed ImpClust algorithm achieves a significantly high score for Silhouette Index, DI-Index, and CH-Index whereas for COP-Index and DB-Index the proposed ImpClust algorithm achieves a significantly low score in comparison to the five existing clustering techniques.

73

Authors: Kapil Kumar, Abhishek Kumar Mishra

This research study proposes a comprehensive approach to improve text recognition in natural photographs. The approach addresses challenges such as curved text, low-resolution images, and efficient recognition algorithms by integrating rectification, visual feature extraction, semantic context modeling, and global context modeling. The model comprises several modules for text recognition. The rectification module normalizes non-uniform text areas using a spatial transformation network to recognize curved text accurately. Visual feature extraction captures complicated patterns and improves picture discrimination with ResNet50. The global context module addresses text sequence dependencies, whereas the semantic context module gathers semantic information. Transformer-based text decoding uses masked multi-head attention. Evaluating benchmark datasets (TotalText, CTW1500, and ICDR15) demonstrates promising results, with the ResNet50 backbone achieving impressive F-measures of 89.3%, 86.4%, and 93.7%, respectively. This research successfully combines rectification, visual feature extraction, semantic context modeling, and global context modeling techniques to address challenges in text recognition for natural photographs. The proposed strategies and components contribute to the model's overall improvement of text recognition.

93

Authors: N. Girubagari, T. N. Ravi

Perpetual improvements are happening in communication networks, both in hardware and software. This constant improvement provides better communication speeds and improves the user experience. The concept of smart cities is a boon that evolves towards the complete automation of the city and the responsible conservation of natural resources. Modern communication technologies support the gradual growth of smart cities simultaneously; the changes in the network architectures open a gateway to attackers, making the cyber network vulnerable. Manual construction and enhancement of network security schemes and protocols are tedious, time-consuming processes that may not be applicable in real time. This dynamic non-deterministic problem can be solved by combining the combination of Artificial Intelligence techniques that can do marvels in anomaly detection in a typical cyber network behaviour. This work, named parallel ABILSTM and CBIGRU ensemble network intrusion detection system (PACENIDS), is intended to use an ensemble of Altered Bi-directional Long Short-Term Memory (ABILSTM) and Customized Bi-directional Gated Recurrent Unit (CBIGRU) to improve the detection of real-time network intrusion attempts with more accuracy and to prevent the network from intimidating attacks on time. The parallel operational nature of the proposed algorithm ensures a swifter performance towards attack detection. This paper uses an impact based fuzzy feature selection algorithm to improve the performance of the proposed approach. The NSL-KDD dataset is used to evaluate the suggested approach. The proposed PACENIDS achieves 96.59% for binary classification, 94.47% and 97.67% for multiclass without and with feature selection, respectively. The experimental result shows that the suggested ensemble approach increases accuracy and precision and reduces the false alarm rate in the target intruder detection system.

118

Authors: Kavita D. Hanabaratti, Rudragoud Patil

In web service composition, the combination of multiple individual services to create complex workflows introduces additional challenges, and task scheduling is a crucial aspect that significantly impacts the overall system performance. The main challenge in web service composition task scheduling arises from the varying nature of services, which may have diverse processing requirements, data dependencies, and execution times. Hence, to address these challenges, the existing work have proposed many methods. But, as cloud environments often host these web services, the dynamic nature of cloud resources and the unpredictability of workloads add another layer of complexity to the task scheduling process. Due to this the existing works have attempted very less work on providing an efficient replanning task based on real-time changes in resource availability. Hence, this paper proposes a novel approach to cloud workload task scheduling in web service composition that incorporates efficient replanning and time complexity analysis called as Web service composition-efficient re-planning (WSC-ERP). The proposed approach takes into account various factors, such as task dependencies, and resource availability, to generate an optimized schedule that minimizes execution time and maximizes resource utilization. The WSC-ERP has been evaluated using the Montage scientific workload. The results show that the WSC-ERP provides better performance in terms of executional time, power sum, power average, energy consumption and reliability when compared with the Energy-Minimized Scheduling (EMS), and Evolutionary Computing based Web Service Composition (EC-WSC). The results show that the WSC-ERP has showed an improvement of 80.27% and 78.45% for average execution time, 89.98% and 87.49% for average power sum, 63.44% and 41.93% for average power average, 92.58% and 87.87% for average energy consumption, 4.97% and 4.07% for average reliability when compared with the existing EMS and EC-WSC models respectively.

140

Authors: Gunita Arun Chandok, V. Arul Mary Rexy, H. Anwer Basha, H. Selvi

Bankruptcy prediction is the process of measuring the possibility of a company facing bankruptcy or financial distress in the future. An accurate bankruptcy prediction model is valuable for creditors, investors, and financial institutions to assess credit risk, make informed investment decisions, and take appropriate risk management measures. Various methods have been built to contemplate bankruptcy, involving more advanced machine learning (ML) methods and traditional statistical techniques. Typically, this method utilizes financial ratios, accounting data, market performance indicators, and other related variables as input features for predicting the probability of bankruptcy. There has been a growing interest in leveraging the power of neural networks for anticipating the bankruptcy with the emergence of deep learning (DL) methods. With this motivation, this article introduces a new white shark optimizer with deep learning-based bankruptcy prediction for financial risk assessment (WSODL-BPFCA) technique. The presented WSODL-BPFCA technique utilizes a hyperparameter-tuned DL model to predict the existence of bankruptcy. To obtain this, the WSODL-BPFCA technique utilizes min-max normalization to transform the input data into a uniform format. For bankruptcy prediction, the WSODL-BPFCA technique introduces an attention-based long short-term memory (ALSTM) approach. Lastly, the hyperparameter tuning of the ALSTM model was carried out by employing of WSO approach. To exhibit the enhanced performance of the WSODL-BPFCA technique, a widespread set of simulations were performed. The comprehensive comparison study highlighted the improved results of the WSODL-BPFCA technique as 97.61% in terms of different metrics.

159

Authors: Adalagere Nemirajaiah Naveen Kumar, Madigahalli Mallegowda, Atur Venkateshmurthy Krishna Mohan, Kannughatta Narasimhamurthy Shreenath, Channa Krishna Raju

The energy-efficient and secure allocation of virtual machines (VMs) plays an important role at the data center. As cloud computing continues to expand rapidly and the number of cloud users increases day by day, the issue of high energy consumption in complex cloud data centers has become a significant concern. To address this challenge, the consolidation of virtual machines (VMs) emerges as a crucial strategy for optimizing cloud resources efficiently. In this study, a novel security evaluation method is proposed to assist the model available for the virtualized system. A multi-objective model-based firefly algorithm (FA) and harmony search (HS) algorithm are used for the system configuration in VM migration is proposed to measure the security threats such as denial of service (DoS), distributed denial of service (DDoS) and Man-in-the-middle attack. The proposed method also decreases the power consumption, network usage and resource wastage in virtual machines. The proposed algorithm achieves better results compared to other existing methods by utilizing the number of virtual machine blocks with cost migration. The experimental results shows that the proposed FA+HS delivers the performance metrics such as makespan, execution cost and resource utilization and achieved at the 1000VMS of 950, 0.001 and 62 respectively, which ensures the better results compared with the existing methods such as whale optimization genetic algorithm (WOGA), multi-objective whale optimization algorithm-based differential evolution (M-WODE) and joint task scheduling and virtual machine placement (JTSVMP).

183

Authors: Sudhamathy Ganapathisamy, Valliammal Narayan

Machine learning is a popular technology that continuously supports organizations in their corporate strategies, managerial functions, and team building. There are many areas in which organizations can adopt technologies that will support decision-making. Human resources (HR) have received more attention in recent years as a result of the fact that skilled employees are a significant growth factor and a genuine competitive advantage for the business. Machine learning started to be employed in HR management to help with employee-related decisions after first being introduced to the departments of sales and marketing. The goal is to support decisions that are based on objective analysis of data rather than on subjective factors. In this research, a novel long short-term memory with recurrent neural network (LSTM-RNN) method is implemented for the prediction and classification of employee attrition. To avoid the overfitting issue and make further prediction analysis based on the suitable objective of feature subset selection, the feature selection algorithm termed brownian motion based butterfly optimization algorithm is proposed. The presented model is validated on the international business machines corporation (IBM) HR dataset which consists of 35 features of employees like business travel, age, education, department, daily rate, distance from home, etc. On the dataset of IBM HR, the proposed model achieved accuracy, recall, F score, and precision of 96.68%, 96.62%, 96.62%, 96.64% respectively. The outcomes proved that the proposed method provides better results in the classification of employee attrition.

202

Authors: Nguyen Duc Huy Bui, Thanh Long Duong

Transmission expansion planning (TEP) is a large-scale and complex problem. Recently, with the high penetration of renewable energy sources (RESs) into the power system, solving the TEP problems has become a challenge for many methodologies. In this paper, an improved zebra optimization algorithm (IZOA) is proposed to resolve the TEP problems with minimal total cost. In order to increase exploration ability, a Lévy flight function is added to the proposed IZOA method. In addition, the exploitation strategy is replaced with a function that does not depend on the number of interactions to help the search process not getting stuck in the local optimal values. The IEEE 24-bus RTS system with and without RESs is used to test the effectiveness of the proposed IZOA algorithm. The simulation results of IZOA are compared with the original ZOA and grey wolf optimization (GWO) method. In addition, to prove the robustness of the proposed method, IZOA is also compared with other existing methods, such as decision analysis (DA) and best-guided artificial bee colony algorithm (GABC), branch and bound (B&B), forensic based investigation optimization (FBIO) and symbiotic organisms Search (SOS). The simulation results show that the optimal value obtained by the proposed IZOA algorithm reduces 39.2% compared with the original ZOA, 36.4% for GWO, and 32.1 % for DA. Thus, the proposed algorithm is one of the suitable methods, powerful, and reliable techniques for solving the TEP problem with the penetration of RESs.

228

Authors: Ramya R, Ramamoorthy S

The Internet of Things (IoT) has recently transformed many lives into comfort zones. The IoT concepts and their exponential growth are burning research issues that need more space for processing and monitoring. To meet the explosive growth of IoT data, edge and fog computing is being deployed for better data analysis with minimum computational complexity. The data can be gathered and analyzed at the fog or edge layer to maximize data utilization. This paper presents a novel prediction and resource allocation using Q-based deep extreme neural network learning algorithms suitable for smart healthcare applications. Resource allocation and effective prediction represent challenging missions involving various resources and IoT nodes to achieve effective computations, leading to real-time computational complexity. The proposed system is composed of four modules: (1) the data collection unit (DCU), (2) the data processing unit (DPU), (3) the intelligent prediction module (IPM), and (4) the adaptive resource allocation module (ARAM). The proposed algorithm is trained using the medical training data, which consists of different heart arrhythmias, specifically heart attacks. The proposed algorithm will be tested using the user’s sensing data from the IoT layer to predict the probability of a heart attack and then decide accordingly. The system aims to achieve an improved quality of service (QoS) with less latency. Extensive experimentation results demonstrate that the proposed algorithm has outperformed real-time data monitoring with prediction and resource allocation.

251

Authors: Putu Desiana Wulaning Ayu, Gede Angga Pradipta, Roy Rudolf Huizen, Kadek Eka Sapta W, I Gede Edy Artana

Amniotic fluid is known to play a crucial role in nurturing fetal growth and development during pregnancy by protecting the fetus from shocks, impacts, and pressure on the abdomen of the mother. One of the parameters often examined by obstetricians regarding amniotic fluid is the volume, which should match the pregnancy stage. Investigations focused on identifying amniotic fluid volume has continued to evolve. Previous examinations used image processing techniques for the single deep pocket method implementation but the results achieved were deemed improvable. Therefore, this study aimed to attain higher identification outcomes than previous examinations by using a model comprising a convolutional neural network (CNN) (feature extractor), chi-square (feature selection), safe level synthetic minority over-sampling technique (SMOTE) (data oversampling), and XGBoost (classifier). The proposed model was comprehensively tested with an analysis of feature selection and oversampling effects. Based on a 5-fold cross-validation examination, the proposed model demonstrated superior accuracy performance compared to previous studies, achieving 96.5% accuracy.

278

Authors: Srinivasarao Thumati, S Vadivel, M Venu Gopala Rao

Rapid electric vehicle (EV) adoption has increased the need for public charging infrastructure. Public charging station (PCS) placement and size must be determined to enable efficient and sustainable charging services. This work proposes a novel method for optimal design and allocation of hybrid photovoltaic systems (HPVs) and PCSs in reconfigurable feeders. A novel LFHBA combines the strengths of the levy flight (LF) exploration technique and the honey badger algorithm (HBA) to solve the multi-objective function focused on distribution loss reduction, improve voltage profiles, and improve reliability index. To test the suggested technique, modified IEEE 69-bus distribution feeders are simulated with varied EV load penetrations. LFHBA is compared to basic HBA, butterfly optimization algorithm (BOA), pelican optimization algorithm (POA), pathfinder algorithm (PFA) in computing performance. The comparison analysis shows that LFHBA has lower target values and greater convergence. Reconfigurable feeder topology permits distribution network design changes, improving system dependability and minimizing power losses. In comparison to base case, EV penetration causes to raise the losses by 26.65%, by optimal allocation of PCSs alone causes to reduce the losses by 38.82%, optimal allocation of PVs and ONR causes reduce losses by 92.19%, whereas, simultaneous allocation of PCSs, and HPVs results to reduce losses by 96.47%. This study emphasizes the need of optimizing PCS placement and capacity with HPVs for efficient and sustainable charging services. As shown by its greater performance over other optimization methods, the LFHBA algorithm helps achieve these goals. Reconfigurable feeders and renewable energy sources improve system dependability and power losses, increasing EV charging infrastructure.

299

Authors: Heba Othman, Mwaffaq Abu AlHija, Mohammad A. Alsharaiah

Malware attacks have become a pressing concern in the domain of computer and communication systems, posing significant threats to data security and privacy. A practical approach is represented aiming to enhance the malware detection process with the help of Honeypot. A hybrid model of particle swarm optimization (PSO) integrated with Fuzzy_KNN algorithms is used in this research. Numerical simulations and mathematical analysis are conducted after developing numeric codes of this scheme. The performance and practicality are examined via these evaluation metrics including accuracy, precision, recall, and F1-score. Based on the numerical investigations, the findings confirmed satisfying performance measures of the hybrid model. The FuzzyKNN algorithm does attain the most remarkable effectiveness, achieving accuracy between 99.95% and 99.97%. This model employs a premium method of neighbourhood voting with an element of fuzziness and excels in large or complex datasets where patterns may emerge based on instance similarity.

331

Authors: Attarid K. Ahmed, Huthaifa Al-Khazraji, Safanah M. Raafat

Time delay is a challenging problem that is frequently encountered in many fields of engineering systems. In this paper, an optimal proportional integral minus proportional derivative (PI-PD) controller with a modified smith predictor (MSP) for varying time delay systems is proposed. The optimal value of the tuning parameter sets of the PI-PD controller for step reference tracking and disturbance rejection is obtained based on two swarm intelligence techniques named flower pollination algorithm (FPA) and heap-based optimization (HBO). Three-time delay processes are used for evaluation. To evaluate the performance of each control structure, a compromised cost function between minimizing the control effort and improving the time response of the system is used. The simulation results based on MATLAB show the PI-PD controller with MSP tuned by HBO exhibits better performance than other controller structures.

356

Authors: Mahdi S. Almhanna, Ahmed M. Al-Salih, Tariq A. Murshedi, Rafah M. Almuttairi

This paper highlights the importance of efficient task distribution and robust fault tolerance in network systems. It emphasizes the limitations of relying on a fixed resource quantity and proposes task replication as a solution to improve data availability. The introduced algorithm dynamically determines the optimal number of replicas based on network history, response time, and joint probability of successful servers, aiming to minimize task failure rates. The algorithm's advancements in grid scheduling lie in optimal resource management, fault-aware job placement, adaptability to changing conditions, and efficient fault tolerance through redundancy planning. The algorithm outperforms three other algorithms, showcasing significant enhancements.

383

Authors: Seema Swamy Gowda, Ambika Ramalingappa

Underwater wireless sensor networks (UWSNs) are the kind of wireless sensor networks which is deployed in an underwater environment with the help of physical sensors. UWSN has applications in different fields like the management of disasters, and navigating the underwater environment and marine species. The nodes present in the UWSN do not have an inbuilt battery, so the available energy sources must be used effectively. To retain energy efficiency in UWSN, effective clustering and routing protocols must be utilized in the UWSN environment. But, the cluster-based routing approaches are effective only for conventional wireless sensor networks. Moreover, the problem related to void hole occurs from sensor node and enhances energy dissipation which helps to minimize the life span of UWSN. So, the research proposed multi-objective sand cat swarm optimization (M-SCSO) for the selection of optimistic cluster heads (CH) and provides a better routing path for the transmission of data without a void hole problem. The parameters such as distance between neighbouring nodes, distance between base station and cluster head, residual energy, and node degree are considered while selecting optimistic CHs using MSCSO. Secondly, energy-efficient routing is processed using MSCSO to deliver data packets in the shortest path. The results obtained through experimental validation represent effectiveness of proposed method. The proposed approach obtained a better packet delivery rate (PDR) of 99.32% whereas the existing emperor penguin optimized Q learning achieved PDR of 90%.

403

Authors: Pratik Kanani, Anil Vasoya, Kamal Shah, Neel Kothari, Nilesh Patil, Gayatri Pandya, Mamta Padole

The healthcare industry relies on efficient and fast decision making. This paper aims to expand the Fog computing and Distributed computing domains to optimize quality of service (QoS) in order to facilitate IoT based healthcare applications with low latency requirements and developing a smart fog gateway equipped with an optimized fog algorithm. The purpose of this study is to optimize real-time healthcare data processing using Fog computing, ensuring dependable, rapid decision-making while minimizing delays caused by data transmission and computation. This is also known as Health-as-a-service (HaaS). We conduct an electrocardiography (ECG) analysis utilizing three computing paradigms: Cloud computing, Fog computing, and a heterogeneous distributed Fog computing setup employing the dynamic OptiFog algorithm. This algorithm effectively manages computational resources within the distributed Fog environment, utilizing Raspberry Pi clusters to enhance performance during worst-case scenarios. The response time is measured using Short Message Service (SMS). The OptiFog node exhibited a response better than the Fog node and the cloud node. The OptiFog algorithm not only takes into account different computing parameters like number of cores, memory usage, CPU utilization and response time of the computing node but also assigns dynamic priorities to these parameters to get the best possible processing available. Based on the workload of the task/node, it dynamically decides the job size to save the network bandwidth and to reduce the network overhead. In conclusion, the proposed work demonstrates that optimizing Fog computing with the dynamic OptiFog algorithm is an effective approach to meet low-latency requirements in IoT-based healthcare applications making it a valuable addition to the Health-as-a-Service (Haas) framework for real-time healthcare data processing.

428

Authors: Aria Hendrawan, Rahmat Gernowo, Oky Dwi Nurhayati, Christine Dewi

The number of vehicles traveling between cities has increased significantly with the acceleration of urbanization. This has resulted in several traffic-related issues, including traffic congestion and the need to collect information on the number and variety of vehicles on the road. In this study, we propose you only look once (YOLO) artificial real-time intelligent analysis (ARIA) based intelligent traffic monitoring system. YOLO is an algorithm that is capable of detecting objects in images and recordings. We improved YOLO’s feature extraction capabilities to improve its vehicle detection accuracy. In addition, we proposed detection models with C3X (convolution neural network) module in the backbone of YOLO. In our experiments, the proposed system attained 99,1% accuracy with 98,3% precision and 98,4% recall on datasets obtained from the CCTV monitoring portal of the semarang city government. In addition, the proposed system has a higher average precision than other vehicle detection and classification methods. Considering the current environmental conditions, the proposed system can classify vehicles in real-time. This makes it a valuable planning and traffic-management instrument.

457

Authors: Abbas Issa Jabbooree, Leyli Mohammad Khanli, Pedram Salehpour, Shahin Pourbahrami

Facial expression recognition (FER) effectively enhances human-computer interaction, robot interfaces, and emotion-aware smart agent systems. Most existing FER algorithms focus solely on extracting facial features from the pixels of the face, disregarding the relative geometric positions that depend on facial landmark points. However, these approaches need to be revised regarding accuracy and robustness. This paper introduces a novel approach to FER that combines convolutional neural network (CNN) and support vector machine (SVM) techniques to extract hybrid features, thereby enhancing discriminative power. The proposed system employs the β-skeleton undirected graph for improved geometric feature extraction through several pre-processing stages. A 1D-CNN is utilized for training the geometric features while simultaneously training the same image using a 2D-CNN. The resulting feature vectors from both sub-networks are merged for SVM classification. The performance of the proposed system is evaluated on two widely used datasets: the extended Cohn-Kanade (CK+) dataset and the Japanese female facial expression (JAFFE) dataset, which are commonly employed in face recognition research. Experimental results demonstrate the superiority of the proposed system over existing methods, achieving a recognition accuracy of 96.19% on the CK+ dataset and 89.23% on the JAFFE dataset.

483

Authors: Ramya Boopathi, Erode Subramaniam Samundeeswari

Cloud computing is a decentralized platform that efficiently allows user applications to utilize various resources. However, it faces challenges in task scheduling (TS) and load balancing (LB). To address these issues, several metaheuristic algorithms have been developed. One such algorithm is a hybrid TS algorithm that uses long short-term memory (LSTM) to determine task runtime reliability, tuna swarm optimization (TSO) to schedule optimal tasks with high expected runtime, and the VIKOR technique to backfill remaining tasks. Despite these efforts, the TS among nodes is unbalanced, leading to overloaded or under-loaded Physical Machines (PMs) and high energy consumption. To tackle these problems, this article proposes a hybrid TSLB algorithm to achieve balanced energy utilization and load fairness among PMs in heterogeneous cloud networks. The TSO algorithm is used to select optimal tasks and virtual machines (VMs) for migration to suitable PMs. The selection of optimal tasks is based on the fitness function of all VMs, while the selection of optimal VMs for migration is determined by the fitness function of all PMs in the network. This approach ensures the best mapping correlation between selected tasks and VMs, resulting in effective load distribution. Simulation results show that the LSTM-TSLBTSO-VIKOR algorithm achieves a makespan of 2700 seconds, a mean resource utilization ratio (RUR) of 0.97, a degree of imbalance (DoI) of 0.03, a throughput of 0.86 tasks/sec, memory usage of 26.9MB, a bandwidth of 187MBps, energy consumption of 95KWh, one VM migration, and a load fairness of 1.23 for 1000 tasks. These results outperform the LSTM-TSTSO-VIKOR, CMODLB, APDPSO, LBPSGORA, and GWOLB algorithms.

510

Authors: Marwan Kadhim Mohammed Al-Shammari, Elaf Ayyed Jebur, Halah Hasan Mahmoud, Israa Ibraheem Al_Barazanchi, Ravi Sekhar, Pritesh Shah

The exponential growth of online information has necessitated effective solutions to combat information overload and optimize network resources. Recommender systems (RS) have emerged as critical tools in this context, opening new avenues for research. However, RS encounters formidable challenges in understanding user behavior and preferences, reducing redundancy in recommendations within social networks (SN), and ensuring scalability and accuracy. To address these issues, this study introduces a novel approach that harnesses the power of two neural networks: Bidirectional long short-term memory (BILSTM) for SN behavior analysis and graph neural network (GNN) for modelling consumer behaviour, both represent a powerful Neuroevolution network. The proposed RS, tailored for SNs, demonstrates significant performance enhancements when compared to traditional deep learning and deep reinforcement learning algorithms. The methodology involves a rigorous training process with a 70% training set and 10% validation set to mitigate overfitting, with final evaluation on a previously unseen 20% testing set. Optimization techniques, including momentum and adaptive learning rates, are applied to GNN-BiLSTM, ensuring computational efficiency. The results unequivocally showcase the effectiveness of this approach in generating more precise and contextually relevant recommendations. By leveraging BILSTM and GNN, the RS gains a deeper understanding of user preferences and item relationships, resulting in superior recommendation quality. Performance metrics such as root mean squared error (RMSE) and mean absolute error (MAE) unequivocally demonstrate the superiority of the proposed model over traditional deep learning and deep reinforcement learning algorithms. In conclusion, the integration of BILSTM and GNN in RS offers a promising solution to the pressing challenges faced by existing systems. This hybrid approach significantly elevates the accuracy and efficiency of recommendations in SNs, paving the way for valuable insights and potential enhancements in future recommendation systems which depends on Neuroevolution approach.

536

Authors: Dewi Rahmawati, Riyanarto Sarno, Chastine Fatichah

Marching cubes is the most widely used isosurface algorithm for 3D reconstruction. For the case study, this paper used medical data from an MRI of brain images, especially in the corpus callosum (CC) part, and volume data from the stagbeetle dataset. This case study was selected to highlight the clinical importance of 3D image visualization. This study can help by showing solid anatomy shapes and locations, which can direct the location of a brain injury with a small error of less than 1 mm; therefore, it can support and minimize the risk of brain surgery. The case study is part of the brain called the corpus callosum, usually used as a reference for brain surgery. For the input data, this paper used 2D segmentation using deep learning methods to obtain the CC segments. This paper used 120 patients, 80% for training and 20% for testing from national hospitals. This paper found 11 sagittal slices containing the corpus callosum out of 166 slices for each patient. This work presents an improved MC algorithm that adds twenty new rules to the existing one, strengthening the rules for voxel representation and increasing the original marching cubes algorithm's 15 rules to 35. Thus, large holes are covered in the 3D reconstruction model, making it largely solid. The proposed 3D visualization achieved zero open edges for the datasets from the national hospital. The results showed that applying the improved marching cubes algorithm produced a 3D representation with better and more robust results, as evidenced by the presence of more vertices and triangles and the absence of open edges. Advanced marching cubes are a great way to remove open edges.

572

Authors: Guruh Fajar Shidik, Ricardus Anggi Pramunendar, Edi Jaya Kusuma, Galuh Wilujeng Saraswati, Nurul Anisa Sri Winarsih, Muhammad Syaifur Rohman, Filmada Ocky Saputra, Muhammad Naufal, Pulung Nurtantio Andono

Earthquakes, as unpredictable and potentially catastrophic events, have long captured the attention of geophysicists due to their profound impact on communities. The devastating consequences of these events underscore the critical need for early earthquake prediction systems capable of forecasting location, magnitude, and depth. With the rapid advancements in technology, particularly in the fields of data science, earthquake prediction methods have undergone significant evolution. This study introduces a proposed method that modified ReLU (Rectified Linear Unit) called LUTanh (Linear Unit Hyperbolic Tangent) which combine both benefits of ReLU and Tanh activation functions. This study applied and compared the proposed method performance in long short-term memory (LSTM) and Bidirectional LSTM (Bi-LSTM) algorithms for predicting earthquake disaster. Furthermore, this proposed method was tested in architecture of multiple input multiple output (MIMO) principle in predicting the occurrence of earthquake disaster. The evaluation results decisively reveal that the LUTanh applied in Bi-LSTM model, particularly when optimized with the Adam optimizer, consistently outperforms the LSTM counterpart. Error assessments of LUTanh in Bi-LSTM consistently demonstrate lower average error scores compared to the origin ReLU activation function up to 4% of mean absolute error (MAE) and 3% of mean square error (MSE).

596

Authors: Zeena Mustafa, Ekhlas Kadhum Hamza

Light fidelity (Li-Fi) can be defined as a type of wireless technology that sends data via light waves through LED light bulbs. Light fidelity (Li-Fi) offers high-speed data transmission capabilities and a large unlicensed bandwidth, making it a promising technology for the future. However, factors including interference, wall reflection, and blocking may cause the quality of a light fidelity (Li-Fi) channel to vary from one part of a room to another. Another type of wireless communication technology that offers broad coverage and slow transmission rates is wireless fidelity (Wi-Fi). Since the electromagnetic spectrums regarding such two technologies do not overlap, there is a possibility for building a hybrid Wi-Fi and Li-Fi network that provides seamless and high throughput global communication. One wireless fidelity access point (Wi-Fi AP) and four light fidelity access points (Li-Fi Aps) make up the downlink hybrid system we discuss in this work. Finding an access point (AP) assignment method that will increase long-term system throughput at the same time as still guaranteeing users' pleasure and fairness is challenging. Thus, the authors suggest using a reinforcement learning (RL) algorithm. The algorithm aims to balance the load of multiple access points (APs) by considering both the LiFi and Wi-Fi channels. The results obtained using MATLAB code for a hybrid system based on reinforcement learning (RL) and a standard solution set size (SSS) access technology called TDMA. The system was evaluated in terms of throughput and user satisfaction for 5 users. According to the results, the RL-based hybrid system achieved a throughput of up to 210 Mbps and an SSS of 180 Mbps. Additionally, the user satisfaction was reported to be 100%.

624

Authors: M. Muthuselvi

Wireless sensor networks (WSNs) are appealing contemporary trends that are dispersed throughout the natural world to analyse the information they receive. The information can be sensed and transmitted more effectively by the sensor nodes. Numerous energy-efficient routing techniques were proposed in earlier research to speed up the network by reducing energy consumption frequently, the sensor nodes quickly run out of power. Most recent articles discuss various techniques aimed at lowering energy consumption in sensor networks. In this paper a novel Multi-objective energy and distance integrated algorithm (MEDIA-WSN) has been proposed to routing /clustering technique with low distance and energy consumption. The MEDIA-WSN method has two phases namely cluster head phase, and routing phase for providing security and energy efficiency. In the first phase, the cluster head selection has been done by using honey badger algorithm which consider the fitness parameters of the sensor nodes such as energy consumption and network lifetime. In the second phase, the routing has been done by using improved spider monkey optimization algorithm which calculate the fitness parameters such as distance, energy consumption and network lifetime. From the simulation analysis, the MEDIA-WSN model overall energy consumption (210 J), packet loss rate (0.5%), packet delivery ratio (99.6%), and end-to-end delay (10 s). The proposed MEDIA-WSN model improves the overall energy consumption of 60.74%, 81.3%, an 85% than EADCR, MOCRAW, and FCEEC techniques respectively.

645

Authors: M. Devika Rani, V. Sai Geetha Lakshmi, K. J. Jegadish Kumar, K. Muthuvel, P. Muthu Kumar

Many countries have benefited from industrialisation and there is a growing global demand for power. For a power system to operate steadily, securely, and dependably, infrastructure must be upgraded and supply and demand must be balanced. In particular, reactive power (VAr) compensation is crucial for balancing the demand for reactive power from industries and, as a result, to ensure a sufficient voltage profile and voltage stability in low-voltage distribution lines. This study suggests the use of switched and fixed capacitors for dynamic volt-var controllers to handle heavy industrial loads. The voltage stability improvement, cost reduction, and loss reduction are the three goals of the multi-objective function. a new, straightforward meta-heuristic for war strategy optimisation (WSO) that combines the power loss index (PLI) to narrow the search space and boost the computing effectiveness. For various industrial load growth scenarios, simulations were performed on IEEE 33-bus low-voltage distribution feeder. A comparative study was also conducted using and compared with WOS (i.e., without reduced search space with PLIs) and whale optimization algorithm (WOA). In terms of global optima, the PLI-WSO findings are superior. In basic 33-bus feeder, having 84.78% VAr compensation results in a 34.39 % loss reduction and 33.23 % cost reduction in a 33-bus feeder, whereas having maximum 16% of industrial load growth in 10 years, the losses and costs are increased by 24.41 times to the base case, respectively. However, by having optimal VAr compensation, the losses and cost savings are resulted for 8.627% and 8.404%, respectively. Different load increase scenarios showed a similar type of overall benefit, demonstrating the scalability of the suggested methodology for real-time larger systems.

666

Authors: Ketut Bayu Yogha Bintoro, Tri Kuntoro Priyambodo

The study introduces the learning automata-based AODV (LA-AODV) protocol to enhance vehicle-to-vehicle (V2V) communication in dynamic vehicular Ad-hoc networks (VANETs). Existing routing protocols, such as Ad-hoc on-demand distance vector (AODV) protocols, face significant challenges, including low data transfer rates, higher delay times, lower throughput, and data congestion resulting from rapidly changing network topologies. LA-AODV addresses these issues by optimizing the quality of service (QoS) through the real-time selection of relay nodes based on vehicle speed, distance, and actual position parameters. Simulations were conducted at the Gadjah Mada university (UGM) roundabout in Yogyakarta, Indonesia, using SUMO and NS3 simulators. LA-AODV outperforms AODV with Packet Delivery Ratios ranging from 95% to 99% and Average Throughputs between 36.90 Kbps and 56.50 Kbps. Although LA-AODV exhibits slightly higher End-to-End Delays, it effectively mitigates Packet Loss Ratios ranging from 1% to 4%. These enhancements optimize routing decisions, reduce communication overhead, and enhance network resource utilization.

689

Authors: Taqwa Q. kadhim, Hazim G. Daway, Ahlam M. Kadhim

Improving low-light images is used in many medical, military, and other applications, such as classification, detection, and increasing the clarity of targets. In this study, we improved low-light images by relying on three main techniques. The first is (DCP), the second is color retrieval, and the third is improving the lighting component based on the Lab color space, where the color components are isolated from the light, and then mapping is done using intensity stretch and (adapted histogram equalization). LOL data (containing 485 low-light images) were enhanced, with two types of standards being reference peak signal to noise ratio (PSNR) and non-reference scalas as entropy (EN), average gradient (AG), and natural image quality evaluator (NIQE). By analyzing the results, the proposed method obtained an improvement, and the quality standards reached EN (7.550), AG (9.978), NIQE (3.499), and PSNR (15.013) values. Compared to a number of recent algorithms, which indicates its success in increasing the lighting and contrast in these images.

711

Authors: Ujwala Gaikwad, Kamal Shah

The head and neck squamous cell carcinoma (HNSCC) is a group of malignant tumors that typically originates in squamous cells lining mucous membranes of head and neck regions. This paper proposed a hidden markov random field model (HMRFM) and VGG-16 for effective segmentation and classification of head and neck cancer. The preprocessing is done by using the min-max normalization method which improves the model performance and feeds into the segmentation process. In segmentation, the HMRFM is utilized to effectively identify and isolate segments in sequential data which makes it easier to analyze the accurate data from input sequence. The VGG-16 model is fine-tuned by certain convolutional layers which is utilized due to less image dataset and overcome the overfitting issues. This method is evaluated on the HNSCC-3DCT-RT dataset and attains better results with regards to AUC, accuracy, sensitivity, specificity, and F1-score values about 95.2%, 98.7%, 91.4%, 97.5% and 91.5% correspondingly. The obtained result shows that the HMRFM with VGG-16 is better compared to other techniques like Deep CNN with data augmentation, 3D-unet convolutional neural network (CNN), and voted ensemble machine learning (VEML).

732

Authors: Ippili Saikrishnamacharyulu, Balendra Mouli Marrapu, Vasala Madhavarao

In geotechnical engineering, the stability analysis of reinforced soil slopes is important to ensure the safety and longevity of infrastructures. In this study, a novel multi-objective particle swarm optimized RF-SVR (random forest and support vector regression) model aimed to evaluate the stability of reinforced soil slopes. The proposed Hybrid RF-SVR-MOPSOA model combines the advantages of both machine learning techniques and optimization algorithms and offers enhanced predictive accuracy and efficiency. Assessing the model's effectiveness involves proposed model was compared with three traditional regression models: Elastic net regression (ENR), ridge regression (RR), and lasso regression (LR). Various performance assessment parameters and ROC curve plots were employed to determine the most suitable model for reinforced soil slope stability analysis. The findings indicate that the proposed hybrid RF-SVR-MOPSOA model demonstrates superior performance compared to other traditional regression models. This innovative approach significantly expands the possibilities for enhancing the assessment of slope stability and ensuring safer and more resilient infrastructural development.

771

Authors: Ramakrihnan Shankar, Devarajan Sridhar

Continuous integration (CI) testing is crucial in modern software engineering and test case prioritization (TCP) techniques improve regression testing (RT) by prioritizing test cases (TCs). Various model has been developed to improve TCs failure prediction and prioritization in CI environments. But, prioritizing the TCs on large test suites without loss of information is a major challenging task. To address this, deep reinforcement prioritizer (DeepRP) model is proposed to improve prioritization in TCP on large test suites. This model employs deep reinforcement learning (DRL) model to learn more test case features, such as changes in source code, version control and code coverage. Also, it enhances self-optimization and adaptive ability for TCP. DRL training employs a deep neural network (DNN) structure to approximate various RL functions like value operation, Q function, transformation system and reward function. An RL system called Q-Learning which determines the appropriate action for an agent based on their action-value role. The DeepRP model uses test case features as input data and the priority of the test case as output. The action includes categorising TCs based on given scores, updating evaluations, calculating reward, and storing the chosen score in a temporary vector among the operations. The reward is computed based on the difference among the specified and ideal rankings for improved TCP on large test suites. The actions include sorting TCs based on assigned scores, updating observations, computing a reward and preserving the selected score in a temporary vector. The reward is calculated based on the distance between the assigned and optimal ranks for better TCP on large test suites. Finally, experimental results show DeepRP significantly achieves RMSE values of 0.09, 0.11 and 0.10 on paint control, IOF/ROL and GSDTSR datasets which is lesser than existing models algorithms like Deepgini, Hansie, DeepOrder, LogTCP and RL-TCP models.

794

Authors: Karthick M, Salomi Samsudeen, Likewin Thomas, Priya Darsini V, Prabaakaran K

The internet of things (IoT) integrates plans, operations, data management, and strategies, because they continuously support businesses, they could be a new point of entry for cyberattacks. IoT security is being seriously threatened by viruses and illicit downloads. These dangers run the risk of obtaining sensitive information, damaging their reputation and their finances. The attack prevention in IoT is detected in this work using a hybrid optimisation mechanism and deep learning a frame. A cybersecurity warning system (CWS) is proposed in this paper, by first pre-processing the input, then classifying, and finally optimizing it. With the modified particle swarm optimization algorithm (MPSO), the IDS is more effective at identifying both normal and abnormal connection in the networks. The smart initialization phase combines various pre-processing strategies to ensure that the informational features incorporated in the early development phase have been enhanced. Then the dilated convolutional neural network (di-CNN) is used for classification and is optimized by using MPSO algorithm to detect the attack. The recommended method is implemented in MATLAB stimulator. The effectiveness of the proposed CWS method approach has been assessed utilising performance criteria such as accuracy, precision ratio, F1 score, specificity, and detection rate. According to experimental findings, the proposed CWS technique 99% in which is relatively high compared to existing methods of 78%, 84%, and 90% than TCN-IDS, MCNN and DSBEL.

813

Authors: Wiwik Anggraeni, Moch Farrel Arrizal Kusuma, Edwin Riksakomara, Radityo P. Wibowo, Pujiadi, Surya Sumpeno

In the era of social media and online communication, the surveillance and classification of disease-related information in real-time is crucial. Twitter data on dengue-related are rarely used for classification, especially text-based classification in Indonesian. Even though, the classification of dengue-related news tweets is capable of being utilized for a variety of purposes. This study presents a novel approach to classifying dengue fever-related tweets in the Indonesian context, utilizing the potential of advanced language models and hybrid neural networks. The method proposed incorporates the advantages of two deep learning architectures: Bidirectional encoder representations from transformers indonesian-based (Indo-BERT) and a hybrid convolutional neural network-long short-term memory (CNN-LSTM) model which is still rarely used in this context. Indo-BERT, a pre-trained language model, extracts complex semantic and contextual information from text, enabling a deeper comprehension of dengue-related tweets. The Hybrid CNN-LSTM model processes textual data in tandem, extracting features via convolutional layers while maintaining temporal dependencies. To assure this model's effectiveness in the Indonesian context, a collection of dengue fever-related tweets in Indonesian was created and labeled for supervised learning. Experiments were conducted in multiple scenarios, and the results demonstrated that the combination of the Indo-BERT and Hybrid CNN-LSTM models had superior performance in classifying tweets about dengue illness into five labels, namely infected, awareness, informative, news, and others. The best models deliver an accuracy of 0.91, an F1-score of 0.90, a precision of 0.91, and a recall of 0.89. The hybrid model's efficacy surpasses that of previous approaches with an average difference in accuracy of 0.25, precision of 0.27, and recall of 0.26. It is anticipated that the Health Service can use the results of this classification to improve the dengue surveillance system with the initial data obtained from this classification.

845

Authors: Alaa Imran Al-Muttairi, Mohammed Jabbar Mohammed Ameen, Ali Khalid Jassim

A new single feed, compact size, and wideband circular polarization (CP) antenna is designed in this work. The designed antenna is made of two layers of the commercially available FR-4 substrate. The geometry in this design consists of a proximity-coupled feeding line for excitation and a circular patch with peripheral cutes as a driven patch surrounded by an inverted U-shape stub as a parasitic patch. According to the design idea, every driven and parasitic patch would produce its own CP radiation. Therefore, the antenna's axial ratio bandwidth is expanded by creating two neighboring axial ratio (AR) minima points. An analysis study based on surface current and electrical field distributions is carried out to demonstrate the antenna's operation. The designed antenna is fabricated and measured. Results show that the reflection coefficient bandwidth is (8.5%), starting from (2.33GHz) to (2.55GHz), and the axial ratio bandwidth is (4.5%) starting from (2.42GHz) to (2.53GHz). The obtained co-pol/x-pol ratio is better than 13dB in the antenna broadside and the measured gain is around 4.8 dB over the entire required band. Based on the achieved results, the antenna is a good solution for ISM band applications and MIMO arrays.

869

Authors: Shashikala S, Ravikumar G K

Copy-move image forgery involves digitally modifying an image by copying and pasting content, hiding foreground objects, or highlighting them through duplication. However, as digital image forgery can become an extremely unsafe and challenging technique to classify images effectively, therefore understanding the detection and classification of real and forgery images is essential. In this research, an improved butterfly optimization algorithm-based convolutional neural network (IBOA-CNN) is proposed for copy-move forgery detection (CMFD), enhancing accuracy and convergence speed by expanding the iteration memory. This proposed approach is used to detect and classify images as original or fake accurately and effectively using deep learning (DL). Initially, the image is obtained by the MICC-F220, MICC-F600, MICC-F2000, and CASIA 2.0 datasets and then image pre-processing is performed by converting Red Green Blue (RGB) into a grayscale image. The Local Binary Pattern (LBP), Wavelet Features (DWT), and ResNet-50 are utilized to extract the features from the images, IBOA is used for feature selection and finally, CNN is employed for the classification to classify CMFD as an original or fake image. Existing methods such as Stacked Sparse Denoising Autoencoder- Spotted Hyena Optimizer- Grasshopper Optimization Algorithm (SSDAE-SHO-GOA), Deep CNN using ResNet-101, and CNN are compared with the IBOA-based CNN approach using MICCF-2000 dataset. The proposed IBOA-based CNN achieves a better accuracy of 97.59%, 99.20%, 99.83%, and 98.92% for MICC-F220, MICC-F600, MICC-F2000, and CASIA 2.0 datasets compared with the existing methods like SSDAE-SHO-GOA, Deep CNN using ResNet-101, and CNN.

895

Authors: Henning Titi Ciptaningtyas, Ary Mazharuddin Shiddiqi, Diana Purwitasari, Fuad Dary Rosyadi, Muhammad Nur Fauzan

The optimization of resource allocation methods in cloud computing environments is essential for enhancing system performance and efficiency. The squirrel search algorithm (SSA) is a metaheuristic algorithm based on swarm intelligence, designed to address optimization challenges. One issue that might arises in the SSA is premature convergence. To address this concern, we propose improving the performance of SSA by integrating it with the opposition based learning (OBL) method. The primary objectives of this method encompass the optimization of makespan, throughput, and resource utilization. The improved SSA algorithm was subjected to a comparative analysis with the genetic algorithm (GA), particle swarm optimization (PSO), and the original SSA. The experiment was performed using the CloudSIM simulator, utilizing three different datasets: the SDSC dataset, a simple random dataset, and a stratified random dataset. The factors under consideration for evaluation encompass makespan, average start time, average finish time, average execution time, total wait time, total scheduling length, throughput, resource utilization, total energy consumption, and imbalance degree. From the experimental results, the improved SSA exhibits superiority over other algorithms in the optimization of makespan on a simple random dataset with an average value of 8.333, as well as minimizing overall energy consumption on the san diego super computer (SDSC) blue horizon dataset which has an average value of 449 kWH. The improved SSA exhibits a gradual increase in the experimental results, rendering the outcomes more foreseeable for a greater number of tasks.

13

Authors: M. Mageshwari, R. Naresh

Cloud computing (CC) is a pivotal technology in the entire universe. Cloud platforms allow management applications or administrators with privileged accounts to remotely do privileged activities for particular tasks like eliminating virtual hosts. If private accounts can be leaked and perform a dangerous confidential activity, severe security issues are performed in cloud environments. To sort out these issues, researchers focused on auditing the behaviours of private users. But it was found to be tough to audit fine-grained privileged behaviours automatically for the graphical operating system. Also, it is difficult to prevent users from avoiding the audit mechanism or preventing hackers from attacking the audit method. Therefore, this study presents a new improved sunflower optimization algorithm based encryption with public auditing scheme (ISFOAE-PAS) in the secure cloud environment. The major intention of the ISFOAE-PAS technique lies in the accomplishment of data security, integrity, and auditing in the cloud environment. In addition, the ISFOAE-PAS technique designs a new secret image encryption using the homomorphic encryption (SIE-HE) technique to encrypt the data. Moreover, the ISFO algorithm is derived for optimally choosing the keys related to the SIE-HE technique. Furthermore, the public auditing scheme is designed to improve security in the cloud environment. Finally, the DDoS attack detection process is performed by a deep convolutional autoencoder (DCAE). For demonstrating the greater efficiency of the ISFOAE-PAS technique, a series of simulations were involved. The experimental values inferred that the ISFOAE-PAS technique reaches maximum accuracy of 99.42%.

35

Authors: T. Kumar, R. Ponnusamy

Classification of medical X-Ray image (XRI) is a task where machine learning (ML) approaches are utilized to categorize XRIs into various categories on the basis of some criteria. The core objective is to automate the task of evaluating XRIs and enhance the efficiency and accuracy of diagnosis. Current advancements in deep learning (DL) have given rise to better performance in various medical image analysis processes. Chest radiographs, the frequently performed radiological exam, are specifically significant modalities for which various applications were studied. The issue of publicly available, multiple, large chest X-Ray datasets in recent times has increased investigation. Therefore, this article presents a new barnacles mating optimizer with DL based medical image classification model (BMODL-MICM) for XRIs. The BMODL-MICM technique uses wiener filtering (WF) as a preprocessing step to eradicate the noise. Besides, the ShuffleNetv2 feature extractor is involved to create a collection of feature vectors. Moreover, the BMODL-MICM technique uses the BMO algorithm for the hyperparameter tuning procedure. At last, elman neural network (ENN) technique is applied for XRIs classification. The investigational assessment of the BMODL-MICM approach on the benchmark XRI dataset reported its effectual outcome of 99.67% over the other methods.

60

Authors: Badari Narayan Vijayeendra Srinivasa Murthy, Siddappa Madappa

The growth of data is an ongoing phenomenon, driven by the expansion of cloud networks, the rapid development of computer systems, and the proliferation of highly beneficial applications. To defend networks from various attacks, such as denial-of-service (DoS), SlowHTTPTest, Hulk, Slowloris, and LOIC-type DDoS attacks, the area of intrusion detection has grown significantly. Traditional intrusion detection systems are combined with a variety of security methods to provide security. However, such a system has limitations when it comes to effectively analyzing vast amounts of data. To detect the vast amount of different intrusion attacks and to select the optimum features, an improved marine predators algorithm (IMPA) is combined with a swarm optimization algorithm named particle swarm optimization (PSO) is proposed. The experiments are conducted using the open-source available dataset referred to as the CSE-CIC-IDS2018 dataset. Here, the classification is done using the Bi-LSTM classifier, an improved version of the LSTM classifier. The experimental results showed that the performance of the proposed IMPAPSO-Bi-LSTM model has achieved high accuracy of 99.40%, recall of 98.20% and f1-score of 98.62% compared to the existing techniques such as recurrent convolutional neural network (R-CNN) and federated intrusion detection system integrated with the blockchain (FIDChain).

85

Authors: Snehal Bankatrao Shinde, Kankipati Lahari, Keerthika Chowdary Garimella, Vicharapu Sowmya Sree, Nileshchandra K Pikle, Girish S Bhavekar, Pradnya Borkar, Sagarkumar Badhiye, Mukesh Raghuwanshi

Cardiovascular disease is any illness that makes the heart work less well. Researchers are working on making smart systems that can correctly detect heart conditions from electronic health data using machine learning algorithms. This is because heart conditions can be very serious. In this study, data from patients and key clinical factors are used to identify cardiovascular disease using machine learning. The main goal of the suggested model is to improve the accuracy and reliability of predicting cardiac disease by focusing on parameter tuning, ensemble methods, and recursive feature removal approaches. Our methods for making predictions included logistic regression, decision trees, K-nearest neighbour (KNN), support vector machine (SVM), naive bayes (NB) machine learning (ML) approaches, ensemble technique approaches, and artificial neural networks (ANN) with stress on regularisation. Compared to the other ways, it was found that using a KNN model gave the most accurate results for the model. A number of factors, such as accuracy, precision, memory, and F1-score, were used to judge the models. The KNN model is the most accurate, at 97.8%.

113

Authors: Gomathi R, Logeswari S, Jothimani S, Sangeethaa SN, Arul Sangeetha S, LathaJothi V

Due to its impartiality in emotion identification, micro-expression can be used in emotional computing. Deep learning has proved successful at recognizing nuanced facial emotions. When the stakes are high, micro-expressions (MEs) reveal people's true sentiments. Early micro-expression recognition (MER) methods use traditional traits. MEs are delicate, rapid facial movements, making them harder to measure and annotate than macro-expressions. Recent deep learning (DL) methodologies aim to improve ME and MER performance. Micro-expression in small, localized areas of the face and a lack of large databases now hinder emotional facial movement recognition. To meet these issues, this study suggests a unique attention mechanism dubbed micro-attention that operates in tandem with residual networks called micro expression fusion network (MEFNet). Enhance the ME model by incorporating attention mechanisms to focus on the most informative regions of the face. This can help improve the accuracy of spatial and temporal data extraction, especially during subtle micro-expressions. Extend the two-stream CNN model to incorporate additional modalities, such as audio or textual cues, in addition to the eye and mouth regions. This multi-modal fusion will enable a more comprehensive understanding of emotions and increase the system's robustness to variations in facial expressions. The MEFNET achieved a specificity of 99.2%, sensitivity of 99.5%, and accuracy of 100% on the CAS(ME)2 dataset, and a specificity of 99.6%, sensitivity of 99.3%, and accuracy of 100% on the SMIC dataset. Rendering to the experimental results, the suggested framework compares favourably to the state-of-the-art methods.

134

Authors: Madhu Belur Gopala Gowda, Naveen Kumar Boraiah, Varun Eshappa, GopalaKrishna Chandrashekara

Epilepsy is a neurological disorder that is distinguished by the presence of seizures, which are caused by abnormal brain activity. On average, about 60% of epilepsy cases are due to focal seizures, which affect a specified part of the cerebrum. Epileptic seizures can be cured at the initial stage, but untreated seizure for a long time cause severe effect and sometimes leads to death. Therefore, accurate classification of the epileptic Electroencephalogram (EEG) signal is an essential step in the diagnosis of epilepsy, aiding physicians in making the diagnosis. This research presents an effective classification of epileptic EEG signals using the improved atomic search optimization (IASO) algorithm and the random search strategy (RSS). The IASO is used to select the appropriate features but has a high probability to enhance the convergence rate. However, due to a lack of population variety and exploratory capabilities and this convergence rate may decrease in some circumstances. As a result, the RSS is offered to enhance the solution and improvise the exploration capabilities. The raw data of EEG signals are obtained from TUH, BONN, and BERN datasets. The raw data is pre-processed using an adaptive filtering method. The IASO is utilized in selecting the relevant features which ease the process of classification using LSTM. The results show that IASO attained better classification accuracy of 98.40% for TUH dataset which is comparatively lower than the existing Deep network model with 97%.

158

Authors: Hendri Darmawan, Mike Yuliana, Moch. Zen Samsono Hadi

Identifying pests and diseases is vital for rice farming, but the lack of experts often limits it. We propose a novel method that uses deep metric learning and k-NN classifier to classify pests and diseases in paddy plants. We experimented using different distance metrics, latent dimensions, base models, and optimizers to train the neural model that produces latent representation from images. Then, we used k-NN retrieval approach to find and label similar images from the latent space. During the inference phase, we manipulated the original image using various transformations and fused them with its latent representation to enhance the quality of latent space. The results show that our method surpasses the conventional deep learning classification (softmax classifier). Specifically, our method achieved maximum accuracy of 0.920 on ResNet-50 and 0.878 on ResNet-152. In comparison, the softmax classifier only achieved maximum accuracy of 0.752 on the same modeling scheme. Our method can produce more discriminative and robust data representations for classification tasks. Moreover, latent fusion from input augmentation during inference can also improve accuracy up to 19.2%. We also deployed the best model from our experiment on serverless cloud computing, allowing users to use the platform for prediction and monitoring through GIS.

185

Authors: Endi Rizal Ferdiansyah, Adian Fatchur Rochim, Wahyul Amien Syafei

The rapid growth of mobile internet usage, particularly in LTE networks, poses challenges for mobile operators in maintaining quality of service (QoS) and optimizing network design planning. Accurate traffic volume forecasting is crucial for network planning and effective resource allocation. This study aims to propose an enhancement to the accuracy of the holt's winter multiplicative seasonal (HWMS) method by applying the rolling forecast technique to achieve better forecasting results. Using a public dataset of 56 cells, the evaluation based on mean absolute percentage error (MAPE) reveals the following sequence of prediction errors: HWMS & rolling forecast (20.47%), HWMS only (30.06%), FbProphet (30.45%), and auto regressive integrated moving average (ARIMA) (31.52%). The cells are categorized as "Good" (59%), "Reasonable" (39%), and "Poor" (2%). The results indicate that the proposed method achieves a significantly higher percentage of cells in the "Good" category, with a 45% difference compared to HWMS without rolling forecast, which only achieved 14%. Moreover, the proposed method outperforms ARIMA by 50% and FbProphet by 37% in the same category. Furthermore, when applied to real data from a telecommunications company in Indonesia, the proposed method identified 11 cells that require solutions out of a total of 100 cells. In comparison, the ARIMA method identified 3 cells, FbProphet identified 12 cells, and HWMS without Rolling Forecast identified 9 cells. Thus, the company can provide solutions for the identified 11 cells without the need for excessive investment while still maintaining revenue potential.

213

Authors: Diana Purwitasari, Abid Famasya Abdillah, Safitri Juanita, I Ketut Eddy Purnama, Mauridhi Hery Purnomo

More people have gotten used to online health consultations (OHC) because of the COVID-19 pandemic to reassure themselves of their health conditions or seek other treatment options. The OHC system could use named entity recognition (NER), specifically for health-related texts called biomedical NER (BioNER), to filter text entities from posting history to ease users' finding information. The terms of named entities (NEs) could be related to human anatomy that have some inconvenience or terms to find out any symptoms of the disease. However, OHC posts, especially user questions, are often non-formal sentences and even long sentences or have incorrect medical terms since the users are most likely non-trained medical professionals, which may lead to out-of-vocabulary (OOV) problems. Although long short-term memory (LSTM) architecture is known for its advantage in modeling sequential data like text, and even with the bidirectional version of BiLSTM, it has some difficulties handling those long sentences. A transformer model could overcome the problems. Another problem concerns fewer annotated data in low-resource language OHC texts despite data abundance in the corpus crawling from the OHC platform. To augment data training, our process includes a self-training approach as semi-supervised learning in data preparation to improve a BioNER model. In preparation for our BioNER model, this work observes and makes a comparison on the embedding step, whether stacked embedding of BiLSTM-based or fine-tuning of transformer-based and defines filtering pseudo-labels to reduce noise from self-training. Although the empirical experiments utilized OHC texts in Indonesian as a case of low-resource language texts because of our familiarity, the procedures in this work apply to Latin alphabet-based languages. We also observed other biomedical NER model creation and topic modelling for verifying the extracted entities from the resulted BioNER model to validate the procedures. The results indicate that our framework, which includes preparing data from raw texts into labelled data using self-training, with a confidence threshold of 0.85, to create the BioNER model, can give F1 scores of 0.732 and 0.838 for BiLSTM-based and transformer-based models.

237

Authors: Mayar M. Moawad, Magda M. Madbouly, Shawkat K. Guirguis

The internet of things (IoT) has been industrializing in a variety of real-world implementations to improve human life. As IoT becomes more industrialized, a huge portion of data is being generated by various devices. Edge, Fog, and Cloud security is a topical issue associated with data storage, managing, or processing systems. Once an attack occurs, it has irreversible and disastrous effects on the development of IoT. Therefore, many security systems have been proposed and implemented for the sake of system security. So, this study consists of three steps based on deep learning, and blockchain for IoT (DLBIoT) architecture security. BoT-IoT and TON-IoT datasets are utilized to demonstrate the DLBIoT, also, the performance was measured using a variety of metrics. Results indicate that the proposed architecture is more efficient than other systems. DLBIoT architecture reaches an approximately 5 % increase for F1-Score compared to DT, NB, RF, and TP2SF. Also, the proposed architecture has simultaneously displayed an accuracy of 98.9% that increases steadily as the Learning Rate increases. Additionally, DLBIoT architecture may be utilized for real-time IoT applications.

261

Authors: Samer Saeed Issa, Sinan Q. Salih, Yasir Dawood Salman, Faris Hasan Taha

The detection rate of network intrusion detection systems mainly depends on relevant features; however, the selection of attributes or features is considered an issue in NP-hard problems. It is an important step in machine learning and pattern recognition. The major aim of feature selection is to determine the feature subset from the current/existing features that will enhance the learning performance of the algorithms, in terms of accuracy and learning time. This paper proposes a new hybrid filter-wrapper feature selection method that can be used in classification problems. The information gain ratio algorithm (GR) represents the filter feature selection approach, and the black hole algorithm (BHA) represents the wrapper feature selection approach. The comparative analysis of network intrusion detection methods focuses on accuracy and false positive rate. GBA shines with exceptional results: achieving 96.96% accuracy and a mere 0.89% false positive rate. This success can be traced to GBA's improved initialization via the GR technique, which effectively removes irrelevant features. By assigning these features almost zero weights, GBA hones its ability to accurately spot intrusions while drastically reducing false alarms. These standout outcomes underline GBA's superiority over other methods, showcasing its potential as a reliable solution for bolstering network security.

286

Authors: C. Saranya Jothi, D. Umanandhini

Hepatitis is a disease originating from a viral bacterium that can be avoided by the earlier identification of the disease and also with the proper classification. The proposed methodology is split into two phases: pre-processing technique and classification method. The pre-processing methods are incorporated to improve the accuracy and reduce the complexity of data. In the first phase, the pre-processing method such as (i) First, label encoder with one hot encoder (LE-OHE), is adopted to convert all the categorical data into a numerical column (ii) Then, Fuzzy logic with may fly (FL-MF) is developed to fill the missing values and (iii) Finally, inter-quartile range (IQR) technique is used to detect the outliers. In the second phase, the classification method is refined by following three steps (i) A correlation-based decision tree (CDT) is constructed to interpret the data as simple, (ii) A reduced error pruning tree (REPT) technique is employed to split the data using grid search algorithm and (iii) The different combination of REPT is given as input to the ensemble stacking method that enhances the predictive performance. According to the results, our proposed method outperforms with an average accuracy rate of 96.1%, and time complexity in both datasets (hepatitis and indian liver patient record (ILPR)) when compared with the existing algorithms such as min max scalar- support vector machine (MMS-SVM), mean imputation-random forest (MI-RF), gradient boosting (GB), random forest (RF), ensemble-stacking (ES), extreme gradient boosting (XGBoost), and radial basis function-extreme learning machine (RBF-ELM). Moreover, the proposed method also reduces the mean absolute error (MAE) error rate.

312

Authors: Sheela Sadashiviah, Nataraj Kanathur Ramaswamy

In vehicular networks, efficient communication between vehicles and infrastructure relies on the ergodic capacities of V2I links. Meanwhile, the crucial transmission of urgent information, collision avoidance, and improved safety hinges on the ergodic capacities of V2V links. Within this context, the present research endeavors to optimize fundamental parameters within a communication system. The ultimate goal is to achieve peak performance by employing the GSA-BPSO (Gravitational Search Algorithm and Binary Particle Swarm Optimization) optimized neural network approach. The primary objective entails maximizing a weighted sum encompassing three critical components. These components encompass the ergodic capacities of Vehicle-to-Infrastructure (V2I) links, Vehicle-to-Vehicle (V2V) links, and the latency requirements tied to V2V links. The study introduces a time delay threshold for V2V data transmission and leverages the GSA-BPSO optimized neural network to optimize key parameters. This enhances system capacity without compromising link communication. Result analysis, specifically for varying time intervals (0.2 ms to 1.2 ms), reveals insights. The Kuhn-Munkres model exhibits the lowest throughput consistently, implying limitations in handling power variations efficiently. The NN model surpasses Kuhn-Munkres but lags behind the GSA-BPSO optimized NN model. Longer time intervals lead to decreased throughput for all models, indicating interference and channel variations. The optimized NN model maintains consistent performance across time intervals, achieving superior throughput under fixed power. The GSA-BPSO optimized NN model outperforms both NN and Kuhn-Munkres models, highlighting its potential for enhancing system throughput with fixed power settings. This research underscores the efficacy of the optimization technique in wireless communication scenarios.

339

Authors: Kossai Fakir, Jules Alexandre Haba, Chouaib Ennawaoui, Mahmoud El Mouden

To ensure proactive and intelligent control of the electrical power generated in industrial units self-producing energy, the construction of a data-driven model devoted to power prediction is an efficient solution that we have proven in this contribution. The remarkable added value of the paper was articulated around three main pillars: primarily, we have modelled a power station based on Rankine cycle that serves the energy needs of an industrial facility. Secondly, the connection of this plant to the public electric grid has been taken into consideration in modelling. Thirdly, we have used the model to avoid paying unwanted bills thanks to the forecasting of periods when there is an energy importation from grid. Before constructing the model, we have prepared a dataset composed of more than 25 million data points, reflecting the history of 3 years of archived data related to fifteen 15 features, in order to predict the target variable of this study. After finding that GRU and LSTM are among the most successful deep learning algorithms in energy prediction, we decided to use them in our particular application. Afterwards, a comparative study between them was first carried out on accuracy, where encouraging results were obtained on both sides, with respective scores of 98.73% and 99.36% during the training step, followed by 98.51% and 99.11% during the testing phase. Besides, the evaluation in terms of loss functions gave a convergence of the GRU model after 8 epochs, against 11 epochs for the LSTM algorithm. Therefore, we preferred to use LSTM in the definitive model because of its higher accuracy, providing that good computing power is used, so that the processing time could be acceptable by the end-users. Finally, we deployed the model to simulate a real case of forecasting the power drop one week in advance. In this case, we found that if the industrial unit had used the results of the model as a proactive decision support tool, it would have avoided importing 1456.4 MWh for 8 days from the national grid, which was equivalent to a loss of 119,094.8 USD.

366

Authors: Azzeddine Sardi, Abdelhadi Ennajih, Jamal Zbitou, Ahmed Errkik

In this work, a novel power divider/combiner based on microstrip technology is investigated and designed for ultra-wideband wireless applications. The design process adopted for this power divider consists of modifying the planar transformers of the conventional power divider using the open artificial transmission technique. The power divider was designed on the basis of the theoretical equations of the open stub technique. It is simulated and optimized using the ADS and CST Microwave Studio solvers. As the simulation results, over the operating band, a bandwidth of 3000 MHz between 1 GHz and 4 GHz was achieved, along the entire band. This gave a maximum return loss of -28 dB at the center frequency and an insertion loss of about 3.5 dB, while maximum isolation of -38 dB was obtained across the output terminals. On the other hand, a prototype of the proposed power divider, fabricated on an FR4 substrate using microstrip technology, gave results confirming those of the simulation. In addition, the final structure considerably reduced the overall dimensions to 75% less than the traditional power divider. Therefore, the proposed power divider could be a suitable solution for wireless applications including 5G, WiMAX, WIFI, and Bluetooth.

386

Authors: Anass Zaidouni, Mohammed Abdou Janati Idrissi, Adil Bellabdaoui

Information system (IS) and information technology (IT) Portfolio value management enables to measure the intended IS/IT project portfolio value, in pursuit of strategic objectives. Yet, the captured IS/IT project portfolio value should be articulated in the different IS/IT project portfolio segments in order to measure each category value contribution. Moreover, it should consider the strategic objectives variability throughout the portfolio lifespan. Therefore, this study proposes a novel methodology for the strategic portfolio value (SPV) calculation, using the rank order distribution (ROD) weights model for strategic objectives variability modelling, and the value for money (VfM) technique of the management of portfolios (MoPTM) guidance based on expert judgments. Besides, this study provides a new IS/IT project portfolio categorization. An IS/IT project management office (PMO) case study demonstrates the outcome of the new methodology, which is the calculated SPV of 0.68, and how it aids, decision-making and performance management, identify under uncertainty the most strategically valuable IS/IT Project Portfolio categories.

411

Authors: Swetha Pesaru, Naresh K Mallenahalli, B. Vishnu Vardhan

Diabetic retinopathy (DR) is a prevalent and potentially blinding eye disease that affects individuals with diabetes. On the other hand, internet of medical things (IoMT) incorporation into DR grading classification offers a promising future for the field of eye health. It optimizes the diagnostic process, increases accessibility to healthcare services, and ultimately improves patient care and outcomes. Therefore, accurate and timely grading of DR severity is crucial for effective disease management. In this article, DR-grading network (DRG-Net) is proposed, which is a comprehensive approach for DR labels classification using the indian diabetic retinopathy image dataset (IDRiD) dataset. To address the imbalanced nature of the dataset, synthetic minority over-sampling technique (SMOTE) is employed for data balancing, ensuring representative samples for each severity level. Then, transfer learning based residual network-50 (ResNet50) architecture is used to extract features from SMOTE outcome, which is a deep learning model renowned for its ability to learn complex image representations. Finally, graph-based K-nearest neighbours (GKNN) classification which utilizes the spatial relationships between samples to make informed decisions, considering the similarity of retinal images in a graph-based representation is introduced for enhanced grading classification of DR. The simulation results show that, the proposed DRG-Net resulted in improved performance as compared to state-of-the-art approaches such as MSA-ResNetGB, DLCNN-MGWO-VW, OHGCNet, and E-DenseNet BC-121 with an accuracy of 99.93%, and F1-score of 99.85%.

433

Authors: Hussein Mahdi, Nidhal El Abbadi

Glaucoma is a dangerous retina disease that can cause irreversible vision loss. It is called the "Silent Thief of Sight" because it has no symptoms in its early stages. A retinal fluid drainage imbalance causes elevated eye pressure. Over time, this pressure could damage the cells of the optic nerve that transmit visual data to the brain. Consequently, individuals with glaucoma can have progressive vision loss, eventually leading to total blindness if left untreated in the earliest stages. In recent years, there has been a rising interest in utilizing computer vision and machine learning techniques for developing an efficient automated diagnosis system for analyzing the retina fundus images to assist optometrists in diagnosing glaucoma at the early stages and reducing the time, cost, and human errors. All these issues motivate us to suggest an automatic glaucoma screening system as decision support for ophthalmologists at the early stages. The system based on combination decisions comes from three different approaches. Each approach uses different kinds of features for analyzing images. The first approach extracts medical features using the YOLOv8 algorithm to detect the optic disc and cup, then classifies features using the logistic regression model. The second approach extracts texture features using deep wavelet scattering and then employs the long short-term memory layer and sigmoid function to classify features. The third approach extracts deep features using convolutional neural network and then employs the attention layer and sigmoid function to classify features. The proposed system attained average results in testing images for the REFUGE, RIM-ONE-DL, DRISHTI-GS1, and ORIGA datasets, 96.21% accuracy, 98.65% specificity, 83.13% recall, 92% precision, 87.34% F1- score, 97.5% area under curve, and the average time is 0.4 seconds for classifying each image.

458

Authors: Spoorthi Basavarajappa, Shanthi Mahesh

Nodule segmentation in lung computed tomography (CT) images is a significant part of the detection and diagnosis of lung cancer. Automatic analysis of lung CT images is necessary to calculate lung nodule characteristics for recognizing malignancy. In recent years, deep learning and neural networks have been used in medical applications. Deep learning utilizes neural networks to train huge amounts of information which effectively learns the nodule features in lower to higher grades for segmenting and predicting the medical images. In this paper, the proposed cat swarm optimization (CSO) based recurrent neural network (RNN) is utilized for lung nodule segmentation. The proposed model is estimated on the freely accessible lung image database consortium and image database resource initiative (LIDC-IDRI) dataset. The proposed model is segmented by using the markov random field (MRF) based firefly algorithm (FA) and cuckoo search algorithm (CSA). The result shows that the proposed CSO based RNN model delivers performance metrics like dice coefficient (DC) loss and accuracy values of about 96.28% and 90.28% respectively, which ensures accurate nodule segmentation in lung CT images.

482

Authors: Optimal Integration of Photovoltaic Public Charging Stations in Multi-Consumer Electrical Distribution System Using Improved Honey Badger Algorithm

The penetration of electric vehicle (EV) loads in the electrical distribution system (EDS) is anticipated to increase dramatically over the next few years in response to rising global warming and fuel price scenarios. Planning for charging infrastructure and renewable energy sources (RES) is urgently required in this context. By carefully combining public charging stations (PCSs) backed by photovoltaic (PVs) systems and distribution static compensators (DSTATCOMs), this study proposes an optimisation strategy to reduce the adverse effects of EV load penetration in EDS. The proposed multi-objective function aims to reduce the voltage deviation index and distribution losses while considering various operational constraints. By combining the voltage stability index (VSI) and opposition-based learning (OBL) strategy with honey badger algorithm (HBA), the optimisation problem is resolved with a smaller search space for global minima. The effectiveness of the proposed technique on IEEE 33-bus EDS was evaluated in various situations. The real power losses are reduced to 242.35 kW, 75.32 kW and 41.76 kW from 350.25 kW by optimally integrating (i) PCSs alone, (ii) simultaneous PCSs and PVs and (iii) simultaneous PCSs, PVs and DSTATCOMs, respectively. Further, the VSI of the EDS is enhanced 0.7057, 0.8727, 0.9018 from 0.6104, respectively. The computational characteristics of HBA were also measured and compared with those of the COA and HBA. In terms of target and computational time, the outcomes produced by IHBA are very competitive with those of COA and superior to those of HBA. Additionally, even with a high EV load penetration, the suggested methodology produces reduced distribution losses and an appropriate voltage profile in EDS.

505

Authors: Raghavendra L R, Manjunatha R C

Achieving high spectral efficiency is a fundamental objective in cognitive radio (CR) networks, where fast and precise spectrum sensing plays a vital role. The cognitive radios rely on spectrum sensing to detect available frequencies, identify idle channels, and utilize them for transmitting data. A new approach to spectrum sensing (SS) is presented in this study, using a machine learning-based model. The model leverages different techniques for feature extraction and separates redundant and non-redundant features to enhance detection performance and reduce training overhead. A hybrid approach is introduced, combining energy, MME, and cyclostationary features extracted from sample data with NCA. The selected features are then trained with a Bayesian-optimized random forest classifier (RF) as a machine learning model. The proposed algorithms are evaluated in terms of average training time, classification speed, and accuracy. The experimental results demonstrate that the proposed bayesian-optimized random forest classifier (RF) achieved a probability of detection of 0.94.

531

Authors: K. Paul Joshua, D. Srinivasa Rao, D. Madhivadhani, Lily saron grace

Mobile ad hoc networks (MANET) are self-launching and self-building multichip wireless networks in which the system's construction deviates significantly owing to the mobility of the nodes. Network nodes serve as both hosts and routers, transferring data from other network nodes. MANETs necessitate an efficient routing protocol with a quality of service (QoS) component. They do not consider mobility as a potential restriction for routing concerns. It is challenging to maintain the necessary QoS in the network due to node mobility, which frequently causes connection failures and high error rates. The aim of the proposed study is to develop an effective MANET framework with dynamic routing and enhanced clustering. In this work, Fuzzy C means clustering is utilized to form the clusters identifying communities or groups of nodes with similar characteristics or behavior, whereas sea lion optimization (SLnO) is utilized to attain the best routing path for data transmission. The proposed model is empirically tested and compared with particle swarm optimization (PSO), ant colony optimization (ACO), grey wolf optimization (GWO), and genetic algorithm (GA) in terms of factors such as throughput, packet loss, energy consumed and end to end delay. The findings demonstrate that SLnO algorithm outperforms other protocols in terms of efficient routing generating a throughput of 680Mb/s and a packet delivery ratio (PDR) of 93.6%.

553

Authors: Asha Kethaganahalli Hanumanthaiah, Manju More Eshwar Rao, Soumyashree Muralidhar Panchal

Breast cancer is a deadly disease; an accurate and early diagnosis of breast cancer is the most efficient method to decrease the death rate. But, in the early detection and diagnosis of breast cancer, differentiating abnormal tissues is a challenging task. In this paper, a weight-based AdaBoost algorithm is proposed for an effective detection and classification of the breast cancer. An AdaBoost algorithm effectively classifies the breast cancer classes by adding the weights to the samples in the week classifier during the training phase. A weighted vote is performed on the results of each week classifier, and the strong classifier is integrated according to the weight of the week classifier. The breast cancer image datasets named CBIS-DDSM and MIAS are utilized for effective classification. Tumor-like regions (TLRs) are diagnosed by utilizing the optimum method of Otsu thresholding to enhance training abilities. The convolutional neural network (CNN) architectures of the AlexNet and ResNet50 are utilized for the feature extraction. A weight-based AdaBoost algorithm is proposed for the classification of breast cancer mammogram images into four classes benign calcification (BC), malignant calcification (MC), benign mass (BM) as well and malignant mass (MM). The results shows that the proposed weight based AdaBoost algorithm delivers the performance metrics such as accuracy, specificity, sensitivity, precision and F1-score values about 99.56%, 99.38%, 99.40%, 98.89% and 99.18% respectively, which ensures the accurate classification results compared with the existing methods such as IMPA-ResNet50, Gray difference weight and MSER detector, MLO and CC methods.

576

Authors: Amal R. Saleh, Motaz A. Elsaban, Mohamed M. Saleh, Hisham M. AbdelSalam

Real estate tax base assessment and evaluation systems are crucial for financing public services and digital economy transformation. To automate and unify complex tax procedures, an intelligent taxation model is needed. Deep learning neural network (DLNN) models have been limited for small-scale, high-dimensional data, but this paper creates an automated model using DLNN, offering superiority, interpretability, and dependability. The model offers lower error levels and high accuracy. The intelligent automated taxation model will improve real estate tax inspection offices efficiency by enabling precise tax base assessments and valuations. Deep learning techniques may enhance real estate price projections, resulting in more precise assessments of taxes. The findings reveal that, compared to other benchmark price predictors, the proposed model achieves greater accuracy (95%–99%) in different datasets and it has the potential to be generalized for real estate taxation authorities and tax inspection offices. The approach is helpful in automated real estate price prediction and taxation applications. This ground-breaking research study proposes a revolutionary strategy that employs deep learning neural network DLNN and challenges the conventional approaches to real estate tax base assessment.

600

Authors: Vijay Anand Krishnan, Abel Thangaraja Ganesan

Mobile adhoc network (MANET) is a type of wireless configuration that features self-organizing wireless mobile nodes and adaptive network connection. Security is a key concern for MANETs due to their dynamic nature and continually changing topology. To improve security, an adaptive trust threshold-aware secure energy-efficient protocol was designed, which adaptively predicts the threshold value using an artificial neural network (ANN) to evaluate the node’s trust for detecting and preventing the suspected nodes. In contrast, it fails to detect and mitigate conflicting behavior (CB) attacks, in which the suspected node may behave well towards a specific group of nodes and badly towards another group of nodes. Therefore, this article proposes a CB attack prediction using the shared learning-based ANN (CBAP-SLANN) algorithm to predict CB attacks in different nodes as well as different timeslots within the same node. The trust calculation should then take into account the consistency of behavior when employing different node-based observations along with the proposed different time-based observations. Initially, nodes are divided into overlapping clusters, and the trust and various network parameters of each node are observed for every group individually. Then, the ANN algorithm is trained in each group using the observed parameters and the trained model for each group is combined to get the global decision, which helps to predict the CB attack nodes in the network. At last, the simulation outcomes show that the CBAP-SLANN algorithm attains 93.1% accuracy when deploying 300 nodes compared to the trust-based routing algorithms.

620

Authors: Jincy C Mathew, V. Ilango, V Asha

Nowadays, retinal fundus images (RFI) play a crucial role in the early and exact prediction of glaucoma. Glaucoma is a vision-threatening eye illness that primarily impairs the eye nerves, and the serious condition of glaucoma leads to permanent vision loss. Several computer vision-based techniques have been explored in past studies for glaucoma detection, but those have certain misclassification results and are not much focused on early stage detection. This work presents an ensemble deep network (Ensemble DeepNet) model for the early detection of glaucoma via retinal fundus images. Initially, the kernel bilateral filter (kernel BF) is employed to improve the clarity of retinal images and reduces image noise. Then, the brightness of the input image is modified using linear histogram transformation (LHT). Further, the segmentation of the optic disc (OD) and optic cup (OC) is performed using two-stage thresholds U-network (Two-stage TUNet). Finally, a new ensemble DeepNet model is developed based on VGGNet, DenseNet, and CapsNet models using a weighted majority voting technique. The improved aquila optimization (IAO) algorithm is used to find the optimal weights for the ensemble DeepNet model. The proposed ensemble DeepNet model attains the highest of 99.45%, which shows that the proposed ensemble model is highly suitable for predicting glaucoma disease in an early stage.

650

Authors: Nirmala Velusamy, Jayanthi Boopathy

Nowadays, tree-structured deep learning classifier models have been widely used in different applications to ensure effective feature representation and learning. Amongst, dimensional sentiment analysis is the most interactive research field, which intends to identify continuous numerical values in the valence-arousal (VA) space. To achieve this, a tree-structured regional convolutional neural network with long short-term memory (T-CNN-LSTM) model was developed, which predicts the VA ratings of the texts for sentiment analysis. In contrast, the effect of a low prediction rate and difficulty of feature learning in a small number of class samples was not analyzed. Hence, this manuscript proposes an adversarial T-CNN-LSTM (A-T-CNN-LSTM) model for predicting the VA to achieve more fine-grained sentiment analysis. This model develops a semantic-enabled frequency-aware generative adversarial network (SFGAN) to produce more adversarial samples using the generator network and decrease the spectral data loss of the discriminator. It embeds the frequency-aware categorizer (FAC) into the discriminator to determine the input veracity in the spatial and spectral domains. Besides, semantic restricted sampling is employed in SFGAN for synthesizing the image subject to a semantic mask. Further, the created samples are classified by the T-CNN-LSTM for predicting the VA scores of sentences. Finally, the experimental results exhibit that the A-T-CNN-LSTM on stanford sentiment Treebank (SST) and CIFAR-10 databases achieves 90.12% and 91% accuracy than the other tree-structured CNNs.

673

Authors: Purba Daru Kusuma, Anggunmeka Luhur Prasasti

Migration algorithm (MA) and walrus optimization algorithm (WaOA) are two new swarm-based metaheuristics which are first introduced in 2023. As new metaheuristics, the modification of these two metaheuristics is still rare. Based on this circumstance, this work constructs a new metaheuristic called as migrating walrus algorithm (MWA) based on the hybridization of both MA and WaOA to create a better metaheuristic than them. MWA consists of five migrations: four directed migrations and one local migration. The references used in these directed migrations are the best walrus, a randomly picked better walrus, a randomly picked walrus, and two randomly picked walruses. In this work, two assessments are carried out: the comparative assessment and the individual migration assessment. The 23 functions are selected as theoretical use cases. In the comparative assessment, MWA is confronted with five new metaheuristics: MA, WaOA, attack leave optimization (ALO), coati optimization algorithm (COA), and osprey optimization algorithm (OOA). The result shows that MWA is better than ALO, COA, MA, OOA, and WaOA in 19, 19, 19, 17, and 17 functions. On the other hand, the individual migration assessment result indicates that the multiple migration approach is important to maintain the superiority of MWA with the directed migration toward the best walrus becoming the most important contributor. This result also strengthens the necessity of the multiple searches strategy rather than a single strategy.

695

Authors: Ricardus Anggi Pramunendar, Pulung Nurtantio Andono, Guruh Fajar Shidik, Rama Aria Megantara, Dewi Pergiwati, Dwi Puji Prabowo, Yuslena Sari, Lim Way Soong

This study aims to improve the classification accuracy of birdsongs by selecting the most pertinent features. This is important because birds are exceptional environmental regulators, but many species are endangered. The community can be assisted in distinguishing bird species and conserving the local environment if the classification is more precise. Nevertheless, because of disruptive noise and unfavorable qualities in the whispering of these bird species, feature selection focuses on enhancing performance accuracy. The use of the gray wolf optimizer (GWO) technique has been employed to identify the most optimum features from the data after outlier removal by the application of k-means clustering, reducing noise through YAMNet, and feature synthesis using gammatone cepstral coefficients (GFCC). This work utilizes the GWO algorithm to address the constraint management challenges associated with high-dimensional data in birdsong classification. The fitness functions used in this research are derived from the K-nearest neighbors (KNN) algorithm. The objective is to devise innovative ways for effectively managing constraints in the context of high-dimensional data. The number of features was reduced by more than 30.7% compared to the initial number of features and obtained an accuracy of 81.06%, as determined by the evaluation outcomes. This discovery improves precision by 4% and surpasses prior research. In summary, this work showcases the effectiveness of the optimization method, especially of GWO. It makes a valuable contribution to advancing a new workflow for analyzing high-dimensional data, specifically in enhancing the classification of birdsongs.

718

Authors: Anagha Deshpande, Krishna Warhade, Pratap Sanap

Human activity recognition (HAR) has become a highly researched area with numerous practical applications in public safety. Deep learning has revolutionized HAR by introducing novel approaches to tackle its challenges. Abnormal activity recognition enables prompt intervention and enhances public safety. Presently vision-based activity recognition techniques mainly use recurrent neural network (RNN) architectures like LSTM to handle sequential data dependency. However, this approach struggles to capture the spatial information between consecutive frames in video data, limiting their ability to learn spatiotemporal patterns. To address this issue, we introduce a layer called ResAttenConvLSTM2D, a variant of the ConvLSTM layer, and propose a novel architecture for solving the abnormal activity recognition problem. In the residual attention model, attention is applied to the residual connections, enabling the network to concentrate on portions of the input by calculating the attention score at each time iteration during model training. In addition, the proposed approach addresses the challenge of limited resources in handling video data by employing robust key frame extraction methods using an unsupervised K-Means algorithm. The proposed architecture is tested for benchmark datasets, i.e., AIRT Lab, hockey fight, and abnormal human activity with a classification accuracy of 90%, 96%, and 99% respectively, showing comparable accuracy or complexity compared to the state-of-the-art (SOTA) approaches.

740

Authors: Qassim Abd A. Hadi, Ali Saeed D. Alfoudi, Ahmed M. Mahdi

The internet of things (IoT) cybersecurity presents a crucial challenge in our daily lives. An intrusion detection system (IDS) is valuable for protecting IoT data against malicious attacks. Moreover, intrusion datasets are often imbalanced in the number of attacks, increasing the bias in machine learning towards classes with high frequency. Normally, this case affects a training model's performance and ability to make a correct prediction. This paper presents a hybrid model that merges dynamic evolving cauchy clustering (DECS) with ranking classification. The DECS model operates based on the self-similarity principle, strategically distributing data into clusters to counteract the impact of imbalanced data. Furthermore, the rank classification algorithm predicts classes for new attacks. The NF-ToN-IoT dataset was used to test the validity performance of the proposed model and compared with standard machine learning algorithms (K-nearest neighbours, random forest, and decision tree). The proposed model outperforms standard cauchy clustering regarding mean square error (MSE), exhibiting a noteworthy reduction to 0.0534. Furthermore, the silhouette score, which indicates clustering quality, notably improved, reaching 0.4707. Additionally, the proposed model attained an accuracy rate of 67.77% and an F1-score of 76.52%, while the standard Random Forest achieved better accuracy at 66.34% and an F1-score of 68.48%.

771

Authors: Vankayalapati Radhika, B. Sai Chandana

Skin cancer is a prevalent condition that may affect anybody, regardless of age. To lower the fatality rate by catching diseases in their earliest stages, automated skin cancer detection is required. To lower the fatality rate by catching diseases in their earliest stages, automated skin cancer detection is required. In recent years, skin cancer screening and detection have become more popular due to the imaging-based computer-aided diagnosis (CAD) approach. To identify and classify skin lesions, this research developed an automated integrative deep-learning network model. The hair is removed using a dull razor approach, and the noise is removed using an average median filter during image pre-processing. The afflicted lesion regions are found in the dermoscopic images using the Otsu thresholding approach and mathematical morphology. From the segmented lesions, the significant features are extracted using improved capsule network (Improved CapsNet). Finally, the enhanced optimized probabilistic neural network (EOPNN) is used for skin lesion classification. The improved artificial jelly optimization (IAJO) algorithm enhances the EOPNN classifier. A benchmark ISIC dataset is used to validate the proposed EOPNN-IAJO approach. The proposed structure has achieved encouraging results for several metrics with 99.23% specificity, 99.02% sensitivity, and 99.53% accuracy.

799

Authors: Azeddine Loulijat, Mouncef El Marghichi, Mohamed Makhad, Naoufl Ettalabi

The doubly fed induction generator (DFIG) is particularly sensitive to grid faults, representing a significant drawback of this generator type, because of the direct connection of its stator windings to the network. The electromagnetic coupling between the stator and rotor in the DFIG results in an undesirable consequence, where voltage dips lead to excessive stator current, causing high currents in sensitive inverters and overloading the DC-link condenser. This document outlines a comparative study that examines two protection scheme configurations with a proportional-integral (PI) controller for managing transient rotor current and overcharging of the DC-link capacitor in order to optimize the operation of the DFIG during network faults. One conventional design choice combines a crowbar circuit with a DC-chopper, in contrast the other design option proposes integrating a modified protection scheme (MPS) that includes an RpLp parallel circuit with the traditional crowbar. Both schemes are placed between the rotor windings and the rotor-side converter (RSC) to boost the low voltage ride-through (LVRT) capacity of the generator. The comparison of results obtained through MATLAB/SIMULINK simulations reveals the successful performance of these two schemes in reducing elevated intensity current of rotor and DC-link voltage levels in the DFIG machine. Furthermore, the traditional crowbar and MPS circuit design have effectively limited the current intensity of the RSC converter to levels lower than 0.22 kA and 2.95 kA, while absorbing up to 2.53 kA and 1.53 kA respectively. Thus, ensuring that the DC-link tension stays within the safe range when network failures occur. The notable distinction is that the MPS design prevents the decoupling of the rotor from the RSC during a symmetrical voltage dip, enabling effective control of all stator power through the RSC.

824

Authors: Mustafa Yahya Hassan, Ali Hamza Najim, Kareem Ali Malalah Al-sharhanee, Mohammed Ayad Alkhafaji, Ridha Mohammed Alfoudi, Worod Adris Shutnan

The firefly optimized Hybrid CNN-BILSTM architecture is a revolutionary AI-driven method proposed in this study for enhancing resource distribution and optimization, and it represents a significant advancement in the Internet of Things (IoT) network space. The importance of this research is in how it addresses the challenges of the utilization of resources in the rapidly growing IoT environment. Traditional distribution of resources methods usually struggle to capture complex chronological and spatial relationships in IoT networks. Convolutional neural networks (CNN) and bidirectional long short-term memory (BILSTM) systems are thus combined in a new Firefly Optimized Hybrid CNN-BILSTM technique to address these deficiencies. Furthermore, this hybrid structure enables excellent synchronous capture of the spatial patterns and temporal dynamics in the computer model by enabling extensive feature acquisition from IoT network information. The firefly optimisation technique is used to optimize the model's parameters, enhancing the model's effectiveness and resolution. Additionally, IoT networks may use an AI-driven strategy to allocate resources in a way that is intelligent and ecologically friendly, increasing resource usage, efficiency, and waste. By mimicking the seductive behaviour and recreating their illuminating patterns, the firefly optimization approach, which serves as the foundation for the algorithm's optimization process, makes it simpler to determine the most optimum resource allocation arrangement. The proposed AI-driven firefly optimised Hybrid CNN-BILSTM structure outperforms state-of-the-art approaches, as shown by the current study's deep evaluations with an accuracy of 99.88%. It also paves the way for more efficient and intelligent resource management in IoT networks and opens up new research directions for the field of AI-driven optimization for IoT applications.

850

Authors: Suhad Abbas Yassir, Haidar Raad Shakir

Information is an asset that profoundly affects every aspect of our lives and requires protection from malicious attacks. Although storing it in a repository can safeguard this information, techniques to maintain confidentiality during transmission over communication networks are essential. Due to the advent of digital technologies and easy communication channels, most of the information now exists in images or other multimedia formats. To safeguard such formats, it is crucial to encrypt information before transmitting it over a network. Encryption transforms the data such that only the intended recipients can reconstruct it. Image encryption is a widely researched topic and various methods have been proposed. As traditional methods often fail to provide adequate protection, advanced encryption standards are currently being investigated. Chaos-based techniques are strong contenders for cryptography because they ensure confidentiality, nonperiodicity, and randomness and are easy to implement. Therefore, they are frequently utilized and can be combined with other methods to bolster protection. In this study, we developed a hybrid image-encryption technique that leverages hyperchaotic maps and elliptic-curve cryptography to guard data against security vulnerabilities. The proposed method achieves an average entropy information value of 7.99%, indicating its resilience to entropy attacks. Furthermore, it showed a net pixel change rate of over 99% for all the test images acquired from the final diffusion stage. It also recorded peak security values of 0.9897, 0.9848, and 0.9676 in the horizontal, vertical, and diagonal directions, respectively. Overall, the scheme demonstrated a satisfactory performance across nearly all evaluation criteria.

874

Authors: Prema Muthusamy, Gomathi Pudupalayam Murugan

People who are deaf or hard of hearing typically communicate primarily through sign language (SL). A complication arises when a deaf\mute person tries to use SL with someone who aren't familiar with it. The use of modern technological advancements will be useful in this situation. In recent years, prediction of SLs has been greatly improved by deep learning (DL) techniques. Amongst, a spatial-temporal multi-cue (STMC) network combines a spatial multi-cue (SMC) module and temporal multi-cue (TMC) to learn spatial and temporal encoding of multi-cue features like full-frame, hands, face, and stance. While it outperforms single-cue algorithms on large SL datasets, it requires more time to pre-process the dataset. Also, annotating key points is required for the complete training of the STMC network. This article presents a spatio-temporal hybrid cue network (STHCN) using dynamic dense spatio-temporal graph convolutional neural network (DDSTGCNN) and VGG11+1D-CNN+BLSTM feature extractor network to address the fore-mentioned issues of ISL recognition and translation tasks. Initially, the skeleton and full frame data is extracted from the input video. The DDSTGCNN is a combination of dense-GCNN (DGCNN) and dynamic spatial-temporal convolution network module (DSTCNM) which learns spatial features temporal features of skeleton data. VGG11+1D-CNN+BLSTM is used to extract the features from full frame data. Then, the extracted features are fed into a bidirectional long-short term memory (BLSTM) encoder, connectionist temporal classification (CTC), and self-attention based LSTM (SA-LSTM) decoders for sequence learning and inference. Finally, The CTC and SA-LSTM predict ISL from input videos and sentences for ISL detection and translation. Finally, an experiment results reveal that the STHCN model achieves 93.65% accuracy in detecting the ISL.

896

Authors: Sneha G Venkateshalu, Santhosh Laxman Deshpande

The proposed spatial-spectrum sparse aware CNN (SSCNN) approach with a CNN architecture employing sparse aware fusion/compression technique to reconstruct high-quality HSI images, departing from prior approaches. The developed model comprises three phases: creating the dictionary, including sparse coding into the dictionary, and evaluating the quality of the parametric reconstruction for the fusion of spatial data. From fused HSI colors, multiscale features are extracted in the second phase. For classification, spatial, and spectral optimization used, a soft margin decision boundary method to reduce the misclassification error. The regularisation factor C was added to manage the commutation among margin and misclassification sensitivity. SSCNN network enables the training of high-level semantic features of fused bands with the best multi-categorization accuracy of 99.46% and kappa of 99.37% in addition to localization-preserved information within the 143.97 secs elapsed time. The findings of the proposed are helpful in spatial data analysis development in the study area.

925

Authors: V Sai Geetha Lakshmi, M Devika Rani, F Jeno Jasmine, Rathinam Muniraj, P Muthukumar

Power systems around the world are transforming into automated networks for more efficient, secure, reliable, and resilient operations, while satisfying technological, economic, and environmental goals. This study addresses the optimal network reconfiguration (ONR) problem by developing a multi-objective function that targets loss reduction, voltage profile improvement, and reliability enhancement. It recommends merging renewable energy sources (RES) with energy storage systems (ESS) to improve resilience. The complex nonlinear multi-objective, multi-variable, and multi-constraint problem is simplified by an efficient single candidate optimizer (SCO). Further, a novel severity index of interruption (SII) is introduced for reducing the search space and thus, improving the computational efficiency of SCO in the proposed hybridization algorithm. In terms of statistical metrics and computing time, simulations on IEEE test systems show that the SCO-SII approach outperforms the basic SCO, grasshopper optimization algorithm (GOA), cuckoo search algorithm (CSA) and Pathfinder algorithm (PFA) in reconfiguration, RES integration, and robust mode augmentation. The capacity of the methodology to handle uncertainty in current electrical networks emphasises its importance.

946

Authors: Sreenath Marimakalapalli Venkatarama Reddy, Annapurna Dammur, Anand Narasimhamurthy

The accurate and dependable prediction of traffic is essential for the stable and safe implementation system of intelligent transportation. It is difficult to examine both inter-series and intra-series temporal correlations at the same time. However, several recent research in the time domain have attempted to capture both correlations but numerous analyses only capture the correlations of temporal and rely on pre-determined priors as inter-series interactions. In this paper, an attention-based spectral temporal graph neural network (AStemGNN) is proposed for traffic flow prediction and identifying the critical node/link graph network. In the spectrum domain, both the correlations of inter-series and temporal dependency are captured by AStemGNN. It incorporates graph fourier transforms (GFT), which describe the correlations of inter-series, and discrete fourier transform (DFT) which represent temporal dependencies. Following the execution of GFT and DFT, the spectrum representations retain patterns and it was predicted efficiently by the modules of convolution and sequential learning. Furthermore, an AStemGNN automatically identifies the correlations of inter-series data without employing the help of pre-defined priors. The efficacy of the AStemGNN method is demonstrated by employing the PEMS04 and PEMS08 datasets. The existing methods such as attention-based spatial-temporal graph neural networks (ASTGNN), static and dynamic spatial correlation neural networks (SDSCNN), dynamic graph convolutional recurrent imputation networks (DGCRIN), and graph and attentive multi-path convolutional network (GAMCN) are used to justify the efficacy of the AStemGNN. When compared with the existing methods such as ASTGNN, SDSCNN, DGCRIN, and GAMCN, the AStemGNN achieves 0.65%, 1.12% mean absolute error (MAE), 0.78%, 1.05% root mean squared errors (RMSE), and 1.57%, 2.19% mean absolute percentage error (MAPE) in both PEMS04, PEMS08 datasets respectively.

965

Authors: Haidar Raad Shakir

Minimizing the computational volume during cryptography while preserving optimal security has long been a goal of computer scientists. Selective encryption is a scalable trend in multimedia content protection. This study builds on the findings of previous studies. It suggests employing the integer Haar wavelet transform and 3D-chaotic with the discrete fractional random transform to encrypt color images. First, each primary color was wavelet-processed into four sub-bands. The approximation sub-band of the wavelet transformation was encrypted using discrete fractional random. Subsequently, to construct encrypted images, an updated approximation and detailed sub-bands, followed by pixel permutation, were utilized. A secure hash algorithm for a plain image was used to produce encryption keys. Finally, a numerical analysis and robustness evaluations were performed using an encrypted test image. The results demonstrated the high efficiency of the proposed method for image encryption. Additionally, it is secure against a wide range of attacks such as entropy and asymmetric attacks. The proposed solution outperformed the encryption approaches proposed by other researchers in terms of security and performance evaluation when using images. Therefore, it fills the gaps in the literature on state-of-the-art methods.

992

Authors: Raad Farhood Chisab

There are many candidate waveforms for the next-era communication system. Filter bank multi-carrier (FBMC) This system depends on offset quadrature amplitude modulation (OQAM) is one of them. Compared to OFDM, it can offer faster data rates and improved spectrum efficiency. The OFDM has a few disadvantages which as a high PAPR resulting in the requirement for high-power amplifiers. So, to overcome this problem, the FBMC system with OQAM is used. In this manuscript, a proposed method was designed using the modified technique through the slantlet transform (ST) to raise the activity for the system. This scheme is suitable for providing lower losses, higher data rates, and for better spectral efficiency. a proposed method that uses the ST gives a modified performance than the fast fourier transform (FFT). Many parameters were applied to test the activity of the system such As Power spectral density, prototype filters, signal power, velocity, carrier frequency, modulation order, power delay profile, channel doppler model, overlap factor, and efficiency. In all these parameters it can be noticed that the proposed system based on ST has overcome the other systems and gives higher performance.

1021

Authors: Triyanna Widiyaningtyas, Didik Dwi Prasetya, Heru Wahyu Herwanto

Neighbor-based collaborative filtering is one of the prevalent recommendation approaches that apply similarity algorithms. Using a similarity algorithm to improve the accuracy of the recommendation system is one challenge in collaborative filtering. The computation of similarity now heavily relies on user rating and behavior scores. Users' preferences for each product kind (genre) demonstrate the user behavior’s value. The algorithm's weakness is that it only considers genre data when estimating user behavior, regardless of when the user rated the item. Therefore, this study aims to develop a new similarity algorithm by considering the genre and the time weight of the item rating, which is called time loss function-based similarity (TLFSim). Newly assessed items have a higher weight than those estimated for a long time. Our experiment tested the TLFSim’s performance compared to the user-score-probability-collaborative-filtering (UPCF) algorithm using the MovieLens 100k dataset. The experimental results demonstrate that the TLFSim algorithm surpasses the prior approach regarding recommendation accuracy, reducing mean absolute error (MAE) by 8.28% and root mean square error (RMSE) by 4.57%.

11

Authors: R. Ramesh, S. Sathiamoorthy

Diabetic retinopathy (DR) is a diabetic complexity that mainly affects the eye. Generally, an ophthalmologist defines the severity of the retinopathy by directly inspecting colour images and estimating them by visually examining the fundus. Due to the enormous amount of diabetic patients all over the world, it becomes an expensive process. The automated system was designed for accurate recognition of the disease using segmentation and fundus image. The blood vessel segmentation process is used to identify and separate blood vessels from the surrounding tissues in an image. This is a crucial stage in the detection of DR, as it enables physicians to measure and identify changes in blood vessels that are indicative of the disease. Numerous approaches are used for blood vessel segmentation and DR classification, along with deep learning (DL) and machine learning (ML) techniques. In this article, we introduce a novel dandelion optimization algorithm with deep learning based blood vessel segmentation and classification (DOADL-BVSC) model for DR grading. The presented DOADL-BVSC technique involves the concepts of blood vessel segmentation and DL-based classification for DR diagnosis on retinal fundus photographs. To accomplish this, the presented DOADL-BVSC technique uses the fuzzy set type-II approach for the image enhancement process. Next, U-Net with Bi-directional feature pyramid network (U-BFPN) model is exploited to segment the blood vessels in the retinal imaging effectively. Moreover, a squeeze and excitation (SE) network is applied for feature vector generation with DOA based hyperparameter optimizer. Finally, quantum autoencoder (QAE) approach is utilized for identifying and classifying DR into distinct stages. The experimental analysis of the DOADL-BVSC approach is investigated on benchmark DR datasets. The simulation results depicted the superior results of the DOADL-BVSC technique with increased accuracy of 99.93%.

31

Authors: Sripal Reddy Kethireddy, Phani Vidhyadhar Rallapalli, Lokanath Reddy Chilakala, Praveen Kumar Devulapalli

The world has taken a drastic look towards scalar mobile wireless multi media sensor networks (MWMSN) and almost all the nodes (IOT devices) are connected via the internet. The amount of image, video and data transmission has increased significantly. For battery-powered smart gadgets like smartphones and tablets, energy efficiency is a constant issue. To fully utilise the advantages of the multiple-input-multiple-output technology, MWMSN heavily relies on cooperative communication. Since each node in a cooperative mobile multimedia sensor network is mobile, energy consumption and routing present significant issues. Since each node in the network is mobile, energy consumption and routing pose significant problems for cooperative mobile multimedia sensor networks. To address these challenges optimal cooperative relay selection (OCRS) technique was proposed by utilizing velocity, distance, and connection quality factors. The suggested method bases its determination of the link quality for each pair of nodes on the signal strength and distance values. The highest quality of service (QoS) rating, which is determined to ensure route stability, reliability, and durability, is chosen as the relay nodes. Further, we have proposed an energy efficient cooperative multi media transmission (EECMT) to achieve a better compression rate with lower energy consumption. This approach utilizes the auto-correction code feature present in the QR code to further improve the performance of MWMSN. The simulation result shows that the proposed approach gives better performance and saves the energy up to 39% (approximately) compared to the efficient cooperative image transmission (ECIT), energy efficient relay node placement (EERP) and energy aware cooperative image transmission (EACIT) approaches.

68

Authors: Venkat Reddy Kumbam, A. Viji Amutha Mary

Epilepsy is a neurological disorder. It is a problem of abnormality in brain resulting in seizures, unusual behaviours, loss of awareness and sensations. Recently many deep learning-based methods to detect epilepsy from multi lead electroencephalogram (EEG) signals have been proposed. Most of these approaches fused the features extracted in one dimensional or two-dimensional mode from EEG through aggregation without any cross-reference learning. As the result, their performance gain in terms of accuracy is limited and also false positives are higher. Learning the cross-correlation information across multiple leads will solve this problem and there has been no works exploiting cross correlation learning. Existing approaches on usage of multiple modalities are mostly feature fusion or decision fusion approaches without cross correlation learning. This work proposes a cross model attention-based learning to learn enriched features from multiple modalities of EEG with aim to increase the accuracy and reduce the false positives. The performance of the proposed solution was tested against University of Bonn EEG dataset. Due to cross correlation learning from multiple modalities, the accuracy of epilepsy detection in proposed approach has increased to 98% which is atleast 2% higher compared to most recent approaches.

92

Authors: Sholiq Sholiq, Riyanarto Sarno, Endang Siti Astuti, Muhammad Ainul Yaqin

In COCOMO II, the primary input for estimating development effort in person month (PM), duration, and cost is the size of the software. Until now, there are two ways to get the size, namely (1) size is estimated using a line of code of software, and (2) size is estimated using unadjusted function points (UFP), which is one of the functional size measurements (FSM). In this study, we added a new way to obtain the size as the primary input in COCOMO II, namely with COSMIC function points (CFP). CFP has several advantages compared to other FSMs, including UFP. Therefore, like UFP, CFP is converted first to LOC, so the conversion equation must be obtained first. We applied four models to get the conversion functions: Ordinary least squares regression (OLSR), support vector regression (SVR) with linear, polynomial, and Gaussian kernel functions. The four models were applied using a dataset from small-scale business application software in Java. The results showed that PM estimation using the CFP model as the primary input produced better accuracy based on MMRE and Pred (0.25), namely 17%-19% and 67%-80%, than the UFP model on the COCOMO II of 135% and 10%.

113

Authors: B. Nisha, M. Victor Jose

Brain cancer is the 10th leading cause of death for men and women. Magnetic resonance imaging (MRI) based Brain cancer segmentation is challenged by less accuracy and high time consumption, hence, an efficient cancer segmentation method is needed for the medical world. In this research, a novel brain cancer segmentation method on MRI is proposed which is entitled 'Brain cancer object Segmentation using LPSIT method and back propagation network (CS_LPSITBP)'. The own contribution of this paper is the 'Unsupervised segmentation of cancer object using Lanczos-3 interpolation based pyramidal structured iterative thresholding (LPSIT)'. The LPSIT method effectively segments the brain cancer region and delivers the key guidelines to generate the training sample to assist the back propagation neural (BPN) network. Processing of multi size images based pyramidal structure is the novel concept of this paper. A five-stage pyramidal based work is influenced in this paper. True positive rate (TPR) analysis reveals that the proposed CS_LPSITBP method improves the TPR by 1.92% compared to the next best method. Proposed method reaches the time consumption of 16.89 sec, while that of the next best method is only 23.88 sec. Experimental results in terms of FScore, Segmentation accuracy and MCR prove the efficiency of the proposed method compared with existing methods; hence, it can be used as a tool for medical practitioners.

138

Authors: Sujono, Ardyono Priyadi, Margo Pujiantara, Mauridhi Hery Purnomo

This paper discusses optimizing DG location and size to reduce power loss and bus voltage deviation index in a 51-bus radial distribution network. Optimization of DG placement uses the firefly algorithm, and optimization of DG size uses the differential evolution (DE) algorithm. The results showed that the optimal locations were on buses 16, 45, and 15, with the highest sensitivity indices of 0.6667, 0.0612, and 0.0601. DG at a lagging power factor of 0.95 gives optimal results with sizes of 358.5157, 500.0000, and 499.9781 kW. Active and reactive power loss reduction is 44.6948% and 66.7038% of the system’s power loss without DG. In optimizing DG placement, AMFA converges faster than DE, GA, and ICA. In the case of DG size optimization, the DE algorithm converges quicker and gives the most optimal results with a fitness value of 0.12385 which is smaller than the FA, GA, and ICA algorithms.

162

Authors: Hashem B. Jehlol, Loay E. George

Managing big data backups is challenging due to high volumes of redundant data. Data deduplication is widely used but incurs significant computational and time costs. This paper proposes a hybrid deduplication system that combines file-level and block-level methods to enhance deduplication while reducing costs. File-level deduplication eliminates duplicate files, while block-level deduplication is applied to non-duplicated files using a dynamic list of divisors to enhance deduplication. A multi-hash function generates three hash values for each file or chunk to improve chunking speed and reduce hash collisions. The proposed hybrid system outperforms other state-of-the-art methods in terms of time, data deduplication ratio, and deduplication gain. Experimental results show reductions of 97.2%, 91.6%, and 82.1% in data size for Dataset 1, Dataset 2, and Dataset 3, respectively, and demonstrate that the proposed multi-hash function is faster and requires less storage than other hash functions.

185

Authors: Choupiri Shivakeshi, Sreepathi B

Currently, the high-load transmission system uses high energy resources, and due to the utilization of high energy, greenhouse gases are polluted. High load transmission caused more resource usage, causing traffic over the communication channel, and fewer packets were transmitted in more time. So to avoid the issue mentioned above, a novel buffalo-based gossip load balancing routing protocol (BGLBRP). Initially, this research created the desired number of nodes in the software defined network (SDN) environment. Avoid high-load data transmission; the system monitored the transmission with the help of the proposed model. For that, the system identifies the high-load data; it then shares the data into a free hub and performs the load-balancing operation. However, the proposed model needs a lower rate of energy resources for performing the packet transmission operation. The proposed model consumes lower energy about 0.051J, so it reduces greenhouse gas. Moreover, the performance rate of the proposed model was calculated, and a comparison was made through the existing model. In comparison, the proposed model attains a higher parameter rate than the existing models. Consequently, the parameters were measured in terms of node lifetime, communication delay, packet drop, packet delivery rate, and energy consumption. The presented model has recorded a high PDR of 97.5%. Compared to the past works, 10% of PDR has been improved. Subsequently, the communication latency was reduced by 25% than the compared models.

208

Authors: Arif Muntasa, Rima Tri Wahyuningrum, Durotun Nafisah

The purpose of the paper is to classify the Leukemia images. In this paper, we proposed a commutative model of a convolutional neural network for Leukemia image classification. We employ commutative hypercomplex modeling A[+1, -1] and A[-1, +1] to build the new model. We hire an augmentation model to enrich the image data sets for the training sets through rotation, zooming, and flipping. We evaluated our proposed method using acute lymphoblastic leukemia image database type 2 (ALL-IDB2). The results show that our proposed method has delivered the best average accuracy at 96.43% for A[+1,-1] and 97.05 for A[-1,+1]. We have measured and found maximum accuracy at 100% for A[+1,-1] and A[-1,+1]. Comparison results show that our proposed method outperformed K-nearest neighbor, support vector machine-radial basis function, support vector machine-linear, support vector machine-polynomial, naïve bayes gaussian, naïve bayes complement, decision tree, and colour hybrid modelling.

238

Authors: Karthick Natarajan, Kamatchi Annappan

Endometrial cancer (EC) rate is rising progressively worldwide so early diagnosis of EC using medical imaging is vital to increase a patient’s survival rate. To avoid the misdiagnosis of EC on magnetic resonance imaging (MRI) scans by clinicians, an automated staging model based on deep learning has broadly emerged in medical systems. Many deep learning models like convolutional neural networks (CNNs) with transfer learning schemes are developed to classify normal and cancer patients from a sequence of MRI scans. But such models have poor sensitivity to classify EC stages from other benign lesions, resulting inaccurate diagnoses of EC. Also, the classical transfer learning schemes reduce the accuracy of detecting medical images due to the discrepancies in data distribution between the source and target domains. Hence, this article proposes a novel deep EC prediction (DeepECP) model that involves image synthesis and classification processes. First, a feature entanglement generative adversarial network (FE-GAN) is proposed for MRI synthesis that creates a desired MRI sequence according to the complementary features of multiple MRI modalities. Then, a multi-modal CNN with long short-term memory (LSTM) network followed by the fully connected (FC) layer is developed to extract a sequence of cancer features from multi-modal MRI sequences. Moreover, an attention strategy is used to fuse those extracted features and get a final feature vector, which is given to the softmax function to classify EC stages. Finally, the extensive experiments show that the DeepECP model on the TCGA-UCEC and CPTAC-UCEC datasets reaches 93.2% and 93.3% accuracy in detecting EC stages, respectively compared to the support vector machine (SVM), VGGNet-16, InceptionResNet and CNN models.

263

Authors: Jutarut Chaoraingern, Vittaya Tipsuwanporn, Arjin Numsomran

To encourage the good health and well-being sustainable development goal, this article presents the design and implementation of real-time indoor air quality index (AQI) prediction using an artificial internet of things (AIoT) electronic nose integrated into a vacuum cleaner robot. The objective of the proposed method is to implement an effective embedded AIoT solution utilizing sensor fusion and the TinyML framework for the purpose of strengthening the environmental health system with suitable current technology. The high-accuracy sensor outputs of total volatile organic compounds (TVOC), humidity, equivalent carbon dioxide (eCO2), and PM2.5 gathered as the dataset are normalized in the data pre-processing state and utilized to create trained models using dense neural networks (DNN) deep learning algorithms. Tiny machine learning is responsible for neural network training, as it is capable of executing AI algorithms on embedded devices with extremely low power consumption and limited RAM and ROM resources. The testing results demonstrate that the predictive model performed well, with 99% accuracy for a maximum absolute regression error less than 15 and an 18.33 mean square error. The embedded device implementation uses Wio terminals with ARM Cortex-M4F microcontrollers for real-time indoor air quality index prediction and visualization. Experimental results demonstrating the average precision of the indoor AQI prediction were obtained at an average accuracy of over 98% with a computation time of 10 milliseconds and an acceptable usage of ROM and RAM resources of 8.5 KB and 1.1 KB, respectively, along with successive performances that satisfied web application data virtualization using the internet of things (IoT).

289

Authors: Arief Kurniawan, Winnona S. T. R. Gultom, Dion Hayu Fandiantoro, Eko Mulyanto Yuniarno, Eko Setijadi, Mochamad Yusuf, I Ketut Eddy Purnama

Cardiovascular disease (CVD) is dangerous and has a high mortality rate. One of the signs of CVD is arrhythmia. The method to find out the symptoms of arrhythmia is an electrocardiogram (ECG) long-term monitoring. The duration of ECG analysis on long-term monitoring is 6 hours, 12 hours, or 24 hours. The device used for long-term ECG is a Holter device. Arrhythmia analysis has followed: detect QRS complex and continue arrhythmia classification. Holter data is stored in local memory or on the server. Storage on local storage makes it difficult for cardiologists to perform analysis. However, using server storage causes access times to slow if patient data increases. Because of this, the cardiologist needs a powerful processing unit to analyze arrhythmia. This study proposes Arrhythmia analysis in the long-term ECG monitoring system. ECG acquisition is sent and stored in the processing unit. We use a single-board computer (SBC) Raspberry Pi as a processing unit. Besides storing data, SBC is also analyzing arrhythmias. The analysis steps are detecting R-peak using the Pan-Tompkins (PTK) algorithm, removing P and T waves using the Gaussian filter, and arrhythmia classification using the multi-layer perceptron (MLP). MLP is a low computational deep learning, which is suitable for SBC. The total storage delay consists of sending and storing data in the database. In the experiment, the propagation data to the broker is 0.023 s., and the storage time to the SQLite database is 15.16 s. The limited recording time for acquisition data is 21 hours and 36 minutes. The success rate of the device in detecting the QRS complex has a precision (+P) of 98.61% and a sensitivity (Se) of 99.8%. Our classification has good in Acc, Se, Spe, +P, and F1-scores are 99.77, 99.55, 99.55,99.85, and 99.55, respectively. The method is superior to several other arrhythmia classification studies.

313

Authors: Amer Tahseen Abu-Jassar, Hani Attar, Vyacheslav Lyashenko, Ayman Amer, Svitlana Sotnik, Ahmed Solyman

The paper presents a generalized system's structural scheme and components for monitoring and controlling access to robotic systems based on biometric data for decision-making. Mathematical decision-making models for choosing the best alternative based on fuzzy sets are proposed. The fuzzy analytical hierarchy process determined weights of 6 criteria (for k1 criterion weighting factor is equal to 2.2; for k2 criterion – 1.8; for k3 criterion – 0.6; for k4 criterion – 0.6; for k5 criterion – 0.6; for k6 criterion – 0.2, that is, most weighty would be k1 – recognition accuracy) against which the best option was evaluated. Thus, during evaluation of alternative on six criteria it was obtained that t alternative X1 (face recognition in full-face), will have the key value of membership function of resulting fuzzy set alternatives with ideal value of integral criterion, equal to 0.53, and as result, implemented rational choice, taking into account given criteria. The difference is that we have introduced such criteria as anti-spoofing. The access control system for robotic systems works in real-time. It is implemented based on an algorithmic decision-making complex, which includes two-factor authentication to increase security: registration involves entering employee data from the keyboard (password, login, surname, first name, position) and physiological authentication (identification by face). Face detection is proposed to be implemented on several images: face, profile, face, and profile in the mask. A distinctive feature of our development is that often, a main requirement for "access control systems with facial identification" is the frontal location of the face relative to the camera; in our case, the image of the face at an angle was added.

339

Authors: S. Subbalakshmi, K. Madhavi

It is becoming increasingly common to outsource data to the cloud. We recommend that the data owner encrypt his or her data ahead of outsourcing it to the cloud to ensure the confidentiality of the data. However, effective search for encrypted information is an essential and complex issue for cloud data sharing. In the past, searchable encryption (SE) has been proposed in many studies as the most promising solution for the efficient retrieval of encrypted data. However, these mechanisms are open to internal attacks that can be restored by assuming keywords from an open trapdoor (TPD) in an autonomous manner. This raises new security concerns, especially data confidentiality and security of cloud-based shared applications. In this paper, we present an efficient mechanism for the public key encryption method with authentication search (PKE-AS) which securely encrypts data using public-key cryptography and validates recipient authentication before retrieval with a shared secret key (SSK) exchange. The shared key is used by the sender to encrypt the keyword and by the receiver to generate a search TPD. The key idea of the proposal is to make sure that the system is safeguard the user data from intrusive keyword guessing attacks. This contribution provides a secure communication model for secure data sharing, searching and accessing by employing the key exchange protocol among the users. The comparison of PKE-AS with existing methods shows low executing time in analysis with different keyword search with encryption, TPD and Verification. It shows an average of 0.52ms less execution time in comparison results. The analysis of the experiment results demonstrates that the proposed method is suitable for practical use in cloud data sharing.

364

Authors: Veni Chinnasamy, Sridevi Rajasekaran

Cyber security is an essential area of study because of the positive effects that networks have on modern society. As it becomes simpler for cybercriminals to launch novel assaults, the scale and complexity of today's networks continue to grow. Therefore, it is crucial to create an efficient instruction detection system (IDS) capable of constantly monitoring network traffic for suspicious activity and issuing alerts as necessary. Several researchers have employed machine learning (ML) and deep learning (DL) techniques to create an effective IDS. Amongst, an effective IDS detection model convolutional neural network (CNN)- long short term memory (LSTM extracts temporal and geographical aspects of network traffic to lower the false alarm rate (FAR) and raise the detection rate (DR). However, the epistemic uncertainty in intrusion data affect the efficiency of CNN-LSTM model. This paper handle uncertainty information by proposing a multivalued neutrosophic convolutional LSTM (MVN-ConvLSTM) which extracts different deep features. MVN-ConvLSTM uses a neutrosophic set (NS) which considers truth (T), indeterminacy (I), and falsity (F) memberships of every features. First, the intrusion features are mapped to NS domain as six sets, positive degree of truth-membership (T_A), positive degree of falsity membership (F_A), Positive degree of indeterminate towards falsity and indeterminate towards truth membership (I_A), negative degree of truth-membership (T_B), negative degree of falsity membership (I_B), and negative degree of indeterminate towards falsity and indeterminate towards truth membership (I_B). Next, an MVN-ConvLSTM is generated with four parallel paths, two of which take input from T and two from I, and a suitable combination of these paths is used to produce the output. The weights of the neural network are updated in four different directions at once using a back propagation approach. The efficiency of MVN-ConvLSTM to handle uncertainty in intrusion dataset is proved by comparing conventional CNN models which provides lower FAR and high DR. Finally, the test results show that the MVN-ConvLSTM achieves accuracy value of 94.63%, 92.84% and 91.84% on three different datasets CIC-IDS2018, WSN-DS and UNSW-NB15.

386

Authors: Anjaneya P, Rajini G. K

Light field imaging, which can gather global visual data, has garnered attention from the computer world in recent years. Inter and intra frame correlation, big data storage, and poor reconstructed depth quality still plague light field area. In this work, greedy discrete wavelet and poincare recurrence network (GDW-PRN) is proposed for robust lossless image compression of hyper spectral images. First, spatial preprocessing is carried out using discrete reduction wavelet transform to remove unwanted artifacts. Next, dimensionality reduced visual data representation is applied by means of greedy band to the pre-processed images, therefore addressing the issues related to large data storage. Finally, lossless image compression is performed by utilizing poincare recurrence network. The entire proposed work was carried out on light field hyperspectral imagery data using MATLAB simulation tool. The proposed work has been compared to the existing methods such as compression based on differential pulse code modulation (C-DPCM), and 3D wavelet transform and spectrum learning with regression vector (3DWT-SRV). On comparison the acquired findings demonstrate that the suggested GDW-PRN process has high compression proportion light field image storage performance as well as leading to high peak signal to noise ratio (PSNR).

406

Authors: Amel Austine, Suji Pramila Richard Nelson

The proposed system focuses more on maintaining D2D links as long as possible which requires an optimum resource allocation initially. Considering this as the optimization issue, this work finds a solution to solve this by proposing a method based on a hybrid optimization concept with game theory. The hybrid concept includes the hybridization of ant colony optimization and Beluga whale optimization (BWO) with a game theory known as ant Beluga with game theory based optimization (AB-GTO) for resource exchange. This enables frequent updates of interference level indicators of each link and maintains consistent links between devices. The proposed solution will be under the consideration of minimization of interference and maximum sum rate as well. The adopted AB-GTO-based REFD2DC in the LTE-A system employing AB-GTO has a high sum rate value over ACO-Game, T-REX RAN, T-REX DRAN, and T-REX CYC.

428

Authors: Abdalla El-Dhshan, Hegazy Zaher, Naglaa Ragaa

This paper proposes a novel metaheuristic technique called two-stage crow search algorithm (TS-CSA). The TS-CSA extends the original crow search algorithm (CSA), that mimics the intelligent foraging behavior of crows. TS-CSA integrates a two-stage search process to improve the CSA in three aspects. First, the two search stages are designed to balance the algorithm's global search and local search capabilities. Second, TS-CSA incorporates a "leaders group" to enable high-quality individuals to guide other individuals and perform global searches by expanding the search space using the best individuals. Third, TS-CSA proposes a new local search strategy for fine-tuning the individuals of the first stage. The proposed TS-CSA and the original CSA are evaluated alongside other metaheuristic algorithms: Grey wolf optimizer (GWO), marine predators algorithm (MPA), and pelican optimization algorithm (POA) using a well-known benchmark of thirteen optimization functions. These functions are commonly employed to assess optimization algorithms due to their complex landscapes and multiple local optima. The results show that the TS-CSA outperforms all competing algorithms, including the original CSA, in terms of convergence and solution quality across all two categories of optimization functions. Furthermore, the proposed algorithm shows potential for solving complex optimization problems across various domains, making it a valuable tool for practitioners and researchers in related optimization fields.

449

Authors: Zuhair A. Ahmed, Safanah M. Raafat

This paper presents an innovative approach to global path planning (GPP) that uniquely integrates an enhanced Bidirectional A* search algorithm, gradient descent smoothing, and orientation filter interpolation. These methods are applied within a quadratic approximation-based potential field and global cost map. The key idea is successfully integrating these powerful techniques, ensuring minimal impact on computational time while enhancing path size and smoothness. Rigorous testing in environments like the Amazon web services (AWS) RoboMaker Hospital Gazebo environment and the sections at the control and systems engineering building at Iraq’s University of Technology (CSE-UOT) in Iraq shows significant improvement over existing planners like wavefront GPP and robot operating system (ROS) default GPP. The proposed algorithm reduced pathfinding time by approximately 60.3% and path size by 9.0%. These findings highlight the efficiency of the proposed global planner, suggesting it as a robust solution for autonomous robotic navigation in complex architectural scenarios.

473

Authors: Muhanad Abdul Elah Alkhalisy, Saad Hameed Abid

The efficiency of distance learning can be assessed by examining specific student behaviors during online courses and assessments. Detecting abnormal student behavior can guide evaluating the effectiveness of learning. Detection speed, accuracy, and model efficiency are essential considerations in distance learning environments. This study proposes a model for detecting abnormal student behavior based on a modified YOLOv5 object detector. Firstly, to achieve model efficiency, the YOLOv5 feature extraction network is pruned by removing the portion responsible for extracting high-level feature maps of small objects, which consumes significant computing resources and parameters. Also, the model's feature fusion part and prediction layers are adjusted accordingly. A convolutional block attention module (CBAM) is added between the model's neck and prediction head to boost the model's focus on students' areas and detection accuracy. Five behaviors were tested on a recently created dataset. The results show that the suggested SPL-YOLOv5 detection method outperforms the SSD-MobileNet, Faster R-CNN, the original YOLOv5 algorithms and state-of-the-art methods regarding performance. The suggested approach increased mAP@0.5 and mAP@0.5:0.95 by 0.14% and 0.13%, respectively, compared to the original YOLOv5. In addition, the model's inference time decreased by 2.7%, computational resources (FLOPs) by 40%, and parameters by 47%.

501

Authors: Syed.Karimunnisa, Yellamma Pachipala

Cloud computing benchmarks the dream of rendering computing as a utility, providing high agility and reachability from an existing set of technologies. It facilitates a wider dimension to architect and manage remote resources. Cloud technology with exponential growth is drilling towards issues that tend to lower its explored possibilities. As cloud systems by virtue deal with various virtualized resources, scheduling is opted as an important metric for measuring and leveraging performance. But scheduling efficiency is deteriorated by various parameters that pave scope for our research and projects immense need for improvising the overall makespan of the system. The proposed work aims at projecting a greater drift in the first phase by witnessing a sequence of phases like pre-processing the user tasks for improved accuracy, classifying the tasks with respect to resource demand and execution time using the improved density based clustering method (IDCM). The second phase deals with enhanced coot optimization algorithm for task scheduling (ECOA-TS) that proceeds and proves its novelty by adopting Cauchy mutation overcoming the convergence backdrop for generating an optimal mapping between clustered user tasks and VMs. The overall performance of the proposed work overrides by reduced makespan against existing state-of-the-art optimization algorithms like particle swam optimization (PSO), grey wolf optimization (GWO) and whale optimization algorithm (WOA) by 27.41%, 19.8%, and 15.33% respectively.

523

Authors: Arif Nugroho, Eko Mulyanto Yuniarno, Mauridhi Hery Purnomo

Inverse kinematics plays a significant role in the robotic motion control. The flexible way to formulate the inverse kinematics can be achieved by using a neural network. The drawback of the neural network-based inverse kinematics is that it has no feedback mechanism to compensate for the remaining error. To improve its performance, further development should be conducted. In the design of the robotic motion control for the NAO robot arms, this paper proposed a new approach that combines the neural network-based inverse kinematics with the Jacobian. This combination yields a closed-loop control system. This control system utilizes the neural network as the feedforward controller and the Jacobian as the feedback controller. In particular, the neural network-based inverse kinematics has a function to estimate a set of required joint angles for the joint actuators, and the Jacobian function is to compensate for the remaining error of the neural network-based inverse kinematics in the estimation of joint angles. By using this proposed approach, we obtained more accurate joint angles for controlling the joint actuators. The comparison result showed that the averaged MSE for the particle swarm optimization (PSO) was 3.47 x 10-3 rad, 1.19 x 10-3 rad for the neural network, and 3.72 x 10-5 rad for the proposed approach. The performance comparison result indicated that our proposed approach has a lower averaged MSE than the other ones. Accordingly, the result of this research confirmed that our proposed approach can provide more accurate joint angles for controlling the joint actuators such that the robot’s end-effector can be driven along the desired path in the cartesian space.

552

Authors: Adil L. Albukhnefis, Amar A. Sakran, AtaAllah Saleh Mahe, Maryam Imran Mousa, Ahmed Mohsin Mahdi, Aqeel hamza al-fatlawi

Intrusion detection systems are critical in identifying and mitigating cyber threats. However, the intrusion data often contains insufficient features, which can adversely affect the classification accuracy of machine learning algorithms. To effectively select optimal features from intrusion attacks, a highly efficient model is required to extract highly correlated features. Nevertheless, traditional detection systems may experience low accuracy and high false-positive rates. This paper proposes a hybrid model for improving intrusion detection systems using mean-variance mapping optimization (MVMO) and random search (MVMOR). The strategy of the proposed model is MVMO algorithm is used to search for the optimal feature subset. while, the random search is employed to optimize the hyperparameters of the machine learning algorithm (classifier). The proposed hybrid model seeks to optimize the feature selection and parameters of classifiers at some time. The performance of the hybrid model is evaluated on the NSL-KDD as a benchmark dataset. The proposed MVMOR achieves an accuracy of 88%, while the conventional MVMO achieved only an 80% accuracy rate. The empirical results show the proposed model has the potential to offer better protection against various cyber threats, thus making a valuable contribution to the field of Cybersecurity.

576

Authors: Narjes N. Jasem, Sadiq A. Mehdi

This research paper introduces a novel encryption algorithm that utilizes a hyper six-dimensional chaotic system to enhance image security. The algorithm incorporates randomization, switching, diffusion, and XOR operations in a series of stages to ensure robust encryption. Two sets of keys are generated using the chaotic system, adding random and unpredictable elements to enhance the algorithm's security. During the encryption process, the image's color values are transformed into red, green, and blue vectors, enabling precise manipulation at the pixel level. Pseudo-random shuffling, guided by chaotic keys, introduces an additional layer of randomness, making the encrypted image resistant to decryption and tampering attempts. XOR operations further modify the pixel values, while interconnecting the color vectors enhances the encryption's complexity and security. Experimental evaluation of the proposed algorithm demonstrates its efficacy. A high PSNR value of 7.9997 indicates preserved image resolution after encryption. The NPCR value of 99.6002 indicates robust resistance to pixel modifications and the UACI value of 35.3623 showcases uniform pixel distribution. The algorithm also offers a substantial key space of 2947, further strengthening its overall security. Notably, its sensitivity of 1015 highlights its resilience against potential attacks.

598

Authors: Purba Daru Kusuma, Ashri Dinimaharawati

This paper promotes a new swarm-based metaheuristic called four directed search algorithm or FDSA. FDSA is designed as a directed search-based metaheuristic without deploying any neighbourhood search. It contains four directed searches that are carried out sequentially. Each search has its reference. These four references are the best member; the resultant of three shuffled members within the swarm; a shuffled member within the swarm; and the resultant of the best member, a shuffled member within the swarm, and the corresponding member. FDSA implements a strict acceptance procedure so that new solution is accepted only if it provides improvement. The investigation is carried out to evaluate the performance of FDSA in solving the 23 functions. FDSA is also confronted with five new metaheuristics: northern goshawk optimization (NGO), average subtraction-based optimization (ASBO), coati optimization algorithm (COA), mixed leader-based optimization (MLBO), and attack leave optimization (ALO). This work also investigates the contribution or dominance of each search in the context of finding the optimal solution. The result shows that FDSA is superior to all these confronters by consecutively outperforming NGO, ASBO, COA, MLBO, and ALO in the 17, 13, 11, 18, and 11 functions. Its superiority is mainly in the high-dimension functions. Through investigation, there is no dominant search among the four directed searches in FDSA. Meanwhile, multiple search strategy is proven to improve performance significantly. On the other hand, the contribution of neighbourhood search is not significant.

627

Authors: Ahmed Abdulmunem Hussein, Esam Taha Yassen, Ahmed N. Rashid

This research proposes an efficient algorithm for solving the green vehicle routing problem (GVRP), capable of generating high-quality solutions while considering environmental impact and computational efficiency. We employ an adapted grey wolf optimizer (GWO) algorithm and Q-learning (QL) for parameter optimization, introducing a discrete grey wolf optimizer (DGWO), a discrete variant of the GWO. The DGWO leverages the 2-opt technique and the Hamming distance concept, making it suitable for addressing discrete problems like GVRP. The key novelty of our approach is the use of QL to refine the parameters of the DGWO, specifically the number of iterations and number of wolves. This application of QL significantly enhances the efficiency and effectiveness of the algorithm compared to DGWO with manual parameter tuning, highlighting the significance of QL in parameter optimization. The proposed discrete grey wolf optimizer-Q-learning (DGWO-QL) algorithm is extensively validated on benchmark instances of GVRP, demonstrating promising results. For smaller benchmark instances comprising of 20 customers and 3 stations, our approach outperforms in 12 out of 16 instances. When tested on larger benchmark instances, within a range of 111 to 500 customers and 21 stations, it achieves success in 10 out of 12 instances. Compared to existing methods, our approach demonstrates improved performance in terms of solution quality and computational efficiency. The results show the robust performance of the DGWO-QL, particularly under stochastic route scenarios, which underscores the advantage of the proposed technique. This study represents a significant contribution to the current body of literature by underscoring the potential of the DGWO-QL algorithm in generating high-quality solutions for GVRP.

649

Authors: Raghul S, Jeyakumar G

Evolutionary algorithms (EAs) in the repository of the evolutionary computing (EC) paradigm are simplistic and stochastic in nature. The desirable characteristic of adaptive simplicity has sustained research community’s rampant curiosity in EAs. Distributed evolutionary algorithmic (DEA) framework is an instinctive extension of EAs. This paper summarizes the design and implementation of a multi-threaded DEA (MTDEA) framework. Empirical analysis between a traditional DEA framework and a simulated MTDEA is presented. The MTDEA framework was outfitted with three plug-in algorithms to handle commonly occurring faults in distributed environments. The propriety of the MTDEA framework was validated in two customary applications - Image thresholding and Image reproduction. The thresholding and reproduction problems were addressed by incorporation of the differential evolution (DE) and genetic algorithm (GA) methods respectively, into the MTDEA framework. The performances of DE and GA routines were compared with their sequential counterpart frameworks. The MTDEA framework was evaluated for reliability and scalability on a high-performance computing (HPC) system.

677

Authors: Namdeo Baban Badhe, Vinayak Ashok Bharadi, Nupur Giri, Sujata Alegavi

Remote sensing image captioning is the challenging task due to low global information, single feature extraction and lack of detailed image captions. To address these issues, this research proposed a deep attention based DenseNet with visual switch added bidirectional long short-term memory (DADN-BiLSTM) for captioning. In this research, initially the images and captions are collected from captioning dataset to smooth away small structures. After that, a double attention mechanism is applied to DenseNet for capturing weak features and to improve the problem corresponds between image feature and captioning information. At the same time, a clustering-based segmentation is more useful and easier to segment the image as smaller parts to make the access easily. Moreover, a decoder is used to improve the use of captioning context information. Then the proposed system is implemented in PYTHON and the performance is evaluated against existing methods in terms of some relevant evaluation metrics such as, recall-oriented understudy for gisting evaluation, accuracy and bilingual evaluation understudy. Finally, the experimental results achieve higher scores in all evaluation indicators such as 0.8925 BLEU1, 0.8514 BLEU2, 0.8252 BLEU3, 0.8312 BLEU4 and 0.8611 ROUGE score on UCM captions, 0.8532 BLEU1, 0.7912 BLEU2, 0.8351 BLEU3, 0.7215 BLEU4 and 0.8139 ROUGE score on Sydney captions and 0.8125 BLEU1, 0.7501 BLEU2, 0.6812 BLEU3, 0.7254 BLEU4 and 0.8245 ROUGE score on RSICD captions.

702

Authors: Santhosh Kumar S, B.K Raghavendra

Coffee is used by about two-thirds of the population in their daily life in one way or the other. However, their significance is greatly affected by the diseases that occur in the coffee leaves. This work proposes a colour-based filtering technique to determine the diseases that arise in the coffee leaves. Initially, the image obtained from the dataset is pre-processed, and contrast enhancement takes place using top–hat transform-contrast limited adaptive histogram equalization (THT-CLAHE) to enhance the visual characteristics. Next, the background present in an image is removed and the leaf is filtered depending on its colour into healthy and diseased leaves via supremum distance-fuzzy C means (SD-FCM) machine learning technique. Then, to obtain minute details, patches are extracted from the diseased leaf. Afterward, by employing double adaptive weight strategy-mexican axolotl optimization (DAWS-MAO), feature extraction is carried out followed by the selection of crucial features. For the classification of leaf diseases, the selected features are further fed into the tuned hard sigmoidal-multi-layer perceptron neural network (Tuned HS-MLPNN). In the end, based on the area of disease prediction, the severity of the disease thus classified is predicted using machine learning. The proposed Tuned HS-MLPNN method achieves the classification accuracy of 98.65%. Finally, the superiority of the proposed model is analysed with the Support Vector Machine (SVM), Naïve Bayes (NB), and artificial neural network (ANN).

728

Authors: Riyadh Rahef Nuiaa, Saif Ali Abd Alradha Alsaidi, Bahaa Kareem Mohammed, Ali Hakem Alsaeedi, Zaid Abdi Alkareem Alyasseri, Selvakumar Manickam, Mohammed Abdulridha Hussain

In recent years, there has been an increase in distributed reflective denial of service (DRDoS) attacks, particularly those that target open lightweight directory access protocol (LDAP) servers. These attacks involve transmitting a small request to a large number of available LDAP servers, seeking information from all users. Consequently, the servers respond with significantly more data than the original request, amplifying the traffic and overwhelming the target with massive amounts of data. Therefore, this paper proposes a novel model for detecting LDAP-based DRDoS attacks by utilizing an enhanced particle swarm optimization (PSO) algorithm based on an adaptive weighted threshold (AWTPSO) model. The proposed AWTPSO model incorporates network traffic features and LDAP protocol characteristics to identify attack patterns. It further employs an adaptive weighted threshold model to dynamically adjust the threshold value for each feature. The enhanced PSO algorithm optimizes the threshold values, thereby improving the detection accuracy of the proposed model. The proposed AWTPSO detection model has been validated using the recent CICDDoS2019 dataset (LDAP sub-dataset). The experimental results demonstrate that the AWTPSO model effectively detects LDAP-based DRDoS attacks with exceptional accuracy of 99.99% and minimal false positives of 0.01%, surpassing other state-of-the-art techniques. Consequently, the proposed model presents a highly promising and robust solution for detecting the threat of LDAP-based DRDoS attacks on enterprise networks.

748

Authors: Agus Wahyu Widodo, Samingun Handoyo, Irfany Rupiwardani, Yusi Tyroni Mursityo, Imam Nurhadi Purwanto, Heni Kusdarwati

The implementation of a CNN1D model is a big challenge because it involves many hyperparameters which have to be tuned well. The research has goals to compare the classification performance of the C4.5 tree and CNN1D models applied in the imbalanced medical dataset in six performance metrics. The oversampling method is applied to the training dataset to form the oversampling training data which is divided into 5 folds of cross-validation data for tuning hyperparameters of both models. By using the grid search method, there are obtained the optimal CNN1D hyperparameters are [256, 4] for the filter and kernel sizes, and the optimal C4.5 hyperparameters are [25,7] for the minimum number of instances in the splitting node and the depth of C4.5 tree. The models with optimal hyperparameters are trained using the oversampling training data and evaluated their performance in both of original training and testing datasets. In the training dataset, the CNN1D outperforms the C4.5 tree in all six metrics with a value of 99% except in the matthews correlation coefficient (MCC) metric with 73% besides the performance gaps of both models being very large. In the testing dataset, the CNN1D also outperforms the C4.5 tree in four metrics with a value of around 99%, but in the MCC and the area under curve (AUC) metrics, the C4.5 tree outperforms the CNN1D although their performance gaps are narrow enough. An interesting future work is to use the MCC or AUC criteria in the tuning hyperparameters of models for the classification of imbalanced class datasets.