International Journal of Intelligent Engineering and Systems

12

Authors: Shreenath Kannughatta Narasimhamurthy, Vasantha Kumara Mahadevachar, Ram Kumar Thayur Narasimhamurthy

Copy-move forgery detection (CMFD) is a well-known digital image forgery technique that is used in detecting forgery regions of images. The copy-move forgery in images occurs when a specific section of an image is attached to a new component of the same image to replicate the forged image parts as an original. The forgery appears to be realistic because even after being forged in the objective region, the image acquires all the basic qualities of the original image. Some post-processing functions, such as scaling or rotation invariant, noise handling, hybrid image manipulation, etc., limit the capability of CMFD models. To overcome these limitations, the features are extracted from the Hybrid feature extraction using modified speeded up robust feature (M-SURF) with accelerated KAZE (A-KAZE) feature description and detection algorithm. The main advantages of the SURF descriptor are: the SURF features are resistant to several post-processing attacks and use Laplacian of gaussian to differentiate background and foreground features clearly. The A-KAZE descriptor is robust in detecting scale-invariant of forged regions and poorly localized key points of objects. A-KAZE constructs the invariant scaling levels using nonlinear diffusion filtering. It smooths the image while preserving the edges and decreasing speckle noise. As a result, it captures the precise location of features during the feature extraction process. Following the feature extraction, the matched features are compared, and removed false matches by applying random sample consensus (RANSAC) technique. The performance evaluation of the proposed Hybrid MSURF with A-KAZE CMFD method is validated on MICC-F220, MICC-F2000, CoMoFoD datasets and surpasses the existing CMFD methods. The proposed CMFD achieved better results compared to existing methods such as when CMFD based on DHE-SURF features along with mDBSCAN clustering and CMFD based on SURF, BRISK features along with DBSCAN clustering measured in terms of Precision of 99.93%, Accuracy of 99.46%, F1-score of 98.21%, true positive rate (TPR) of 99.93%, and false positive rate (FPR) of 4.67%.

39

Authors: Amir Yasseen Mahdi, Siti Sophiayati Yuhaniz

This paper proposes an improved binary sparrow search algorithm (IBSSA) as a search strategy within the feature selection (FS) methods. Its main objective is to use clinical texts to improve COVID-19 patient categorization. The constant need for an efficient FS system and the favorable outcomes of swarming behavior in numerous optimization situations drove our efforts to develop a novel FS strategy. Additionally, clinical text data are frequently highly dimensional and contain uninformative features, which have a major impact on the classifier's accuracy, making FS a key machine-learning step in data pre-processing to reduce data dimensionality. The bi-stage FS approach is used in this work to elect the features. At the initial stage, we employed a term weighting scheme (TWS) that assigned a weighted score to each feature by measuring the significance of the features obtained from the pre-processing model using a new weight calculation method called root term frequency-core-inverse exponential frequency (RTF-C-IEF). Next, finding the most relevant and almost optimal feature subset for COVID-19 illness diagnosis is done in the second stage using a freshly developed methodology that was inspired by the way sparrow’s behavior. The suggested modification method for the sparrow’s algorithm is composed of several stages of advancement. The main objectives are to promote the exploration of the search space and increase the algorithm's variability. In order to evaluate the proposed model, various classifiers were employed on two datasets, each of which had 1446 and 3053 cases, respectively. The experimental and statistical results demonstrate that the proposed IBSSA is significantly superior compared to other comparative optimization algorithms, and it successfully upgrades the shortcomings of the original SSA. Moreover, the IBSSA has the highest accurate performance when compared to other rivals by the SVM classifier, Where, average removed features are 77.99% and 83.5%, with improvement percentages by F1-scores: 84.95% and 95.94 % for both datasets, respectively.

69

Authors: F. Al-dolaimy, Rabei Raad Ali, Noor Nabeel, Wisam Subhi Al-Dayyeni, Fatima Hashim Abbas, Hussein Muhi Hariz, Salama A Mostafa, Mohammed Ahmed Jubair

Vehicular ad hoc networks (VANETs) recently covered a wide range of intelligent transportation systems (ITSs) and applications. VANETs consist of convinced unique characteristics such as dynamic movement and high speed. Due to these characteristics, link failures occur, and delay and routing overhead is greatly increased, directly affecting the effectiveness, Quality of Service (QoS), and stability of VANETs. To achieve an efficient and reliable network performance, this paper proposes QoS Aware Hybrid Optimization for Improving Path Selection (HOIPS-VANETs) in VANETs. This hybrid optimization combines the Improved ant colony optimization (ACO) and Effective Whale Optimization Algorithm (EWOA). The EWOA algorithm is used for initial optimal path selection, and ACO is used to find the best optimal solution to achieve effective communication in VANETs. This optimization technique is also applied to low-density, medium-density, and high-density scenarios, as it is compared with the earlier methods. During the experimentation process, our findings reveal no noticeable change in the performance of the earlier methods when applied to the low-density and medium-density scenarios. Still, the performance is gradually reduced when applied to the high-density scenario. On the other hand, the performance of the proposed approach is better for all three tested scenarios and demonstrates effective communication for the VANETs.

94

Authors: Mohammed Kdair Abd, Luay G. Ibrahim, Firas M. F. Flaih

Incorporating renewable energy storage systems in power grids has presented significant challenges in maintaining a stable power generation structure and load frequency within interconnected grids. A promising solution to support the reliable and safe operation of the power system is the use of pumped storage units because of their excellent adjustment characteristics, including fast power response and convenient start and stop features. To optimize the performance of the pumped storage system, this article proposes the implementation of a linear-adaptive disturbance rejection controller (LADRC) in conjunction with the segmentation particle swarm optimization (Se-PSO) algorithm for parameter tuning. This article establishes a precise hybrid load frequency control (LFC) model for a two-area interconnected power grid incorporating pumped storage power plants (PSPP). It determines the optimal LADRC parameters for varying load disturbances in different areas of the power grid. The proposed control method systematically simulated pumping and power generation conditions in a PSPP. Simulation results indicate that the proposed method enhances dynamic stability by approximately 85%, offering greater anti-disturbance ability and adaptability than the traditional proportional-integral (PI) controller.

115

Authors: Arundhati Nelli, Rashmi Jogdand

Maximizing resource utilization by minimizing service level agreement (SLA) violation is considered to be extremely difficult process for scheduling of realtime workload on multi-cloud platform. In particular the proposed work considers both energy and performance as SLA minimization constraint for scheduling of workload in multi-cloud platform. The scheduling optimization of SLA minimization is done as single objective function using dragonfly soft computing approach. Further, the scheduling optimization of min-max is done as multi-objective for maximizing resource utilization with minimal SLA violation using dragonfly soft computing approach. The results show that the min-max workload scheduling (MMWS) has improved the resource utilization by 16.02%, 7.92%, and 2.52% respectively for montage workload and 9.92%, 4.07%, and 1.11% for SIPHT workload in comparison to the existing workload-scheduling adaptive-dragonfly algorithm (WS-ADA), reliability multi cloud-scheduler (REL-MC) and service-level agreement-based workload-scheduling (SLA-WS) method respectively. The results show that the MMWS has reduced the SLA violation rate by 81.92%, 69.23%, and 30.11% respectively for montage workload and 84.57%, 61.32%, and 12.62% for SIPHT workload in comparison to the existing WS-ADA, REL-MC and SLA-WS methods respectively. Hence, from the experiment outcomes it is shown that the proposed MMWS technique reduces SLA violation with improved resource efficiency in comparison with WS-ADA, REL-MC and SLA-WS.

137

Authors: Lakhvinder Singh, Dalip

An intelligent information technology-based system offering practical health advice can improve the superiority of lifespan for individuals with chronic illnesses. This study develops a BFCNN (Bees featured CNN) model for the classification or prediction of heart, liver, and kidney diseases. The main objective of the proposed classification model is to accurately diagnose a patient's disease. The research model aims to construct a highly efficient deep-learning system for patients with heart, liver, and kidney diseases. A novel approach for detecting multiple diseases in medical datasets using a BEE-featured CNN model. The BEE-featured CNN model is a modified version of the traditional CNN model that incorporates the behavior of bee swarms to enhance the accuracy of the model. The bee-inspired algorithm used in the model involves the use of multiple artificial bees that simulate the behavior of real bees to search for optimal solutions. The proposed model was trained and tested using a dataset consisting of medical datasets of patients with different diseases. The images were collected from various sources and labeled with the corresponding disease type. The dataset was split into training and validation sets, with 80% of the datasets used for training and 20% for validation. The results of the experiments showed that the proposed BEE-featured CNN model outperformed the traditional CNN model in detecting multiple diseases in medical datasets. The results of the prediction model demonstrate outstanding performance, with an accuracy of 96.7%, specificity of 99%, and Sensitivity of 97%. The research system demonstrates promising results in classification accuracy and holds potential for usage in medical and e-health applications.

164

Authors: Robert Marco, Sharifah Sakinah Syed Ahmad, Sabrina Ahmad

One of the main reasons that hinders making software effort estimation remains a most of the unresolved problem due to the heterogeneous nature of software data with complex structures. In processing nonlinear data, the long short-term memory (LSTM) model is often used for the purpose of solving discriminative and generative problems. Taking into account the fact that the LSTM network have low computational efficiency, due to the need to set a large number of hyperparameters. However, training deep learning models requires expensive work in terms of specifying hyperparameter configurations in the model. The grid search (GS) optimization method is used to find the best hyperparameter values for deep learning networks, which have many different hyperparameters that affect how well the network architecture works. In this paper, we proposed the grid search method as a quick step toward optimizing the parameters of the LSTM model. An empirical study was conducted using five datasets. From our results, we have seen that the LSTM-grid search model consistently performs better across datasets in mean absolute error (MAE) and root mean squared error (RMSE) compare to existing work on using machine learning approach for software effort estimation. The main advantage of training the LSTM show that the grid search finds hyperparameters which results in faster convergent, the possibility of generalization, and better coefficient of determination in estimating software effort.

191

Authors: Asmaa Ftaimi, Tomader Mazri

Recently, much effort and interest have been devoted to developing accurate and reliable blockage prediction systems. The latest research tends to apply artificial intelligence to enhance blockage prediction systems' accuracy, especially when dealing with non-line-of-sight radio signal propagation. However, AI models can carry inherent vulnerabilities that might damage the relevant system. Therefore, this paper aims to examine the effect of adversarial examples on AI-based blockage prediction systems. By performing adversarial example attacks on the AI-based blockage prediction model, we have shown the drastic impact on root mean square error that has arisen from 0.26 to 0.82 after introducing adversarial examples with 0.08 magnitude. Additionally, we have proposed a defensive approach based on encryption techniques to proactively prevent further compromise of the targeted system. The proposed method not only enhances data privacy but also help to prevent adversarial attacks that exploit transmitted data.

215

Authors: Shruti. R. Danve, Manoj S. Nagmode, Shankar B. Deosarkar

The extreme growth in smart mobile users and the demand for higher data rates have paved the way for developing 5G networks. Additionally, increased wireless data traffic implies more power consumption in network equipment. Mobile network service providers are concerned about the energy efficiency (EE) of cellular networks. In this era of advanced wireless communication, Massive multiple input multiple output (MMIMO) technology satisfies the demand for a hyper-connected society. To deal with the problem of increased hardware power consumption and cochannel interference in MMIMO, a quad methodology is proposed to evaluate the performance of a cellular network. This quad package comprises linear precoding (LP), power consumption reduction (PCR), User selection (US), and antenna selection (AS). The proposed quad methodology is used for enhancing the energy efficiency and capacity of a cellular network. Two LP techniques, zero forcing (ZF) and minimum mean squared error (MMSE) are compared to evaluate the EE of MMIMO with antenna selection. It is observed that EE for MMIMO with AS using ZF is three times greater than MMSE. MMIMO system with the suggested quad methodology provides 45% and 20% improvement in capacity and energy efficiency respectively over only the norm US.

239

Authors: Saif Mohammed Al-azzawi, Mohanad A. Deif, Hani Attar, Ayman Amer, Ahmed A. A. Solyman

Due to global lockdown policies implemented against COVID-19, there has been an impact on electricity consumption. Several countries have emphasized the significance of ensuring electricity supply security during the pandemic to maintain the livelihood of people. Accurate forecasting of electricity demand plays a crucial role in ensuring energy security across all nations; accordingly to achieve this objective, this study employs metaheuristics optimization algorithms to enhance the prediction model's operation, such as Support Vector Machine (SVM), K-Nearest Neighbors (KNN), and Random Forest (RF), at an optimized level to minimize errors. Two metaheuristics optimization methods, Particle Swarm Optimization (PSO) and Genetic Algorithms (GA), are utilized. The suggested prediction models are trained using daily power usage data from three US urban regions. In terms of prediction accuracy, the findings show that KNN with PSO surpasses the other models. The COVID-19 pandemic reduced power usage by 20% relative to pre-pandemic levels.

264

Authors: Purba Daru Kusuma, Faisal Candrasyah Hasibuan

This paper introduces a novel swarm-based metaheuristic called swarm magnetic optimizer (SMO). SMO imitates the behaviour of two magnets close to each other: pushing toward or pulling away from each other. This push-pull mechanism is then adopted to become a novel search in SMO. SMO is set up as a swarm of magnets that move autonomously. In the early iteration, the pull strategy is dominant. Meanwhile, it declines during the iteration and is replaced by a push strategy. The determination between these two strategies is calculated stochastically. SMO consists of three sequential phases in every iteration where the corresponding magnet and its reference run a push or pull search in each phase. The reference used in each phase is the global best magnet, a randomly selected magnet, and a randomly generated magnet. This paper evaluates SMO through simulation to find the optimal solution for 23 functions. The performance of SMO is compared with five latest metaheuristics: mixed leader-based optimizer (MLBO), guided pelican algorithm (GPA), zebra optimization algorithm (ZOA), coati optimization algorithm (COA), and clouded leopard optimizer (CLO). The result shows that SMO is better than MLBO, GPA, ZOA, COA, and CLO in consecutively 22, 15, 11, 13, and 14 functions. The superiority of SMO, especially, is in solving high dimension functions. Meanwhile, in the fixed-dimension multimodal functions, SMO is still superior to MLBO but less superior to GPA, ZOA, COA, and CLO.

289

Authors: Kunjan D. Shinde, Vijaya C

The advancement in very large scale integration (VLSI) technology and field programmable gate array’s (FPGA) parallel constructive nature for digital circuits has made the implementation of finite impulse response (FIR) filters increasingly relevant in real-time. In FIR filters the computational complexity increases with the length of the filter; numerous techniques have been developed to design viable architectures for realizing FIR filters. The multiple input multiple output (MIMO) based parallel FIR filter architecture often requires a large area for realization as the level of parallelism and filter order increases; this study presents an architecturally enhanced, novel parallel architecture for FIR filter design which eliminates the resource dependencies on the level of parallelism. The proposed architecture addresses the issue by limiting the number of multipliers required irrespective of the change in the level of parallelism and further modifies the data flow to improve the iteration period in the proposed design. The level of parallelism (L) is fixed to 8 in the presented study as the value is neither too low nor high; it is important to note that the increase in the level of parallelism will increase the number of samples generated at a given time. The presented work is demonstrated using tap 16 FIR filter implemented on FPGA VIRTEX 4 xc4vsx35-10ff668 and VIRTEX 5 XC5VSX95T-1FF1136 platforms, and filter-order 16 FIR filter is implemented on ARTIX 7 xc7a200t-2fbg676. The functionality is verified using Xilinx ISE 14.7 and Matlab 2018 environments. The post-synthesis results and comparative study show the design validation; the proposed architecture outperforms the benchmark and conventional MIMO-based parallel FIR filter architectures with an improvement in total delay by 70% w.r.t. [2] and 93% w.r.t. [3] and 2% w.r.t. [4] as compared in case studies 1 – 3. The reduction in area metric is observed as dipping in the slice requirement by 31%, 47%, and 110% and dipping in the LUT requirement by 58%, 66%, and 65%, further, the delay metric is improved by 49%, 80%, and 67% when compared with MIMO-based parallel architecture of [1] on the same platforms suggested in [2-4] respectively.

318

Authors: Rafael M. J. O. Silva, Attilio Sbrana, Paulo A. L. de Castro, Nei Y. Soma

Deep learning models, despite their potential, often function as “black boxes”, posing significant challenges to interpretability, particularly in sensitive fields such as healthcare and finance. Addressing this issue, we introduce a novel, human-understandable metric aimed at enhancing the interpretability of local interpretable model-agnostic explanations (LIME). Distinct from previous methodologies, this metric is designed to assess the shift in classification probability upon the removal of features (words), thereby providing a unique insight into interpretability. We deploy a convolutional neural network (CNN) for sentiment analysis, interpret predictions utilizing LIME, and evaluate these explanations using three distinct metrics: our proposed metric, a conventional model-based metric, and human evaluations. Through rigorous validation, our metric demonstrated high recall performance, a key indicator of relevant instance retrieval. Results showed worst-case and best-case recalls of 80.29% and 98.19% respectively, against a logistic regression metric for “good” and “excellent” classifications. Comparisons with human evaluations using single-word explanations revealed worst-case and best-case recalls of 82.03% and 94.37%, respectively. These high recall values highlight our metric's effectiveness in aligning with both human judgments and model-based metrics, emphasizing its capacity to capture essential explainability aspects. Furthermore, our study also outlines certain LIME limitations, setting the stage for future interpretability-focused AI research.

348

Authors: Abhishek Singhal, Devendra Kumar Sharma

Voice signals are the most convenient and essential method to establish the communication between human beings. The linguistic characteristics of the human being can be recognized with the help of voice signal analysis such as the ascent of language, health status, emotions, age, gender, etc. Gender is an unprecedented peculiarity of the speakers. This peculiarity can be recognized through the analysis of the voice signals. This paper presents the precision value of gender, i.e., transgender, female and male. The live voice samples for females and males are recorded as well as the live voice samples for the transgender in India are recorded first time because the voice samples of the transgenders are not available at the authenticated sources. The error factor and accuracy is also calculated for the proposed model, which is utilized for the gender identification of the speakers by analyzing the voice signals. In the proposed work, bidirectional long short-term memory (BiLSTM) architecture with recurrent neural network (RNN) model is used as a classification algorithm with mel frequency cepstral coefficients (MFCCs). The MFCC is utilized as a trait of the signals used to identify the gender of the human beings. The outcomes of the proposed model are not dependent on the content of the articulation and the dialect of the speakers. The precision values for female, male and transgender, the accuracy and error rate of the proposed model, depend on applied quantities of voice signals in testing and training datasets. The average precision value is 93.91%, 94.70%, 97.43% and 95.35%for the male, transgender, female and proposed model, respectively. It is found that the female precision value is the highest, and the male has the lowest. Further, the average error rate of the proposed model is 0.045. The gender identification accuracy is 95.52%.

372

Authors: Robert Marco, Sharifah Sakinah Syed Ahmad, Sabrina Ahmad

The uncertainty inherent in the predictive environment of software effort estimation, is due to the limited availability of data information and the expensive costs associated with data collection. Consequently, there is difficulty in making accurate predictions with insufficient software effort estimation data, due to the limited information in the data. The probabilistic deep generative model conditional variational autoencoder (CVAE) is capable of producing synthetic data and modeling complex data distribution characteristics that is similarity to the real Unfortunately, because the method was previously developed for the classification task, the application of the technique to the regression task has received little attention. This study aims to construct a CVAE model combined with the relevance function contained in the re-sampling method. Relevance function method aims to create a label on the target. Statistical tests, such as the Levene test, t-test, and Kolmogrove Smirnov test, are utilized to compare the results to the real data. This study considers seven popular algorithms for comparison, such as synthetic minority oversampling technique regression (SMOTER), generative adversarial networks (GAN), conditional tabular generative adversarial networks (CTGAN), gaussian-copula (GC), copula GAN (CGAN), and variational autoencoder (VAE). CVAE approach has the best augmentation performance quality with accuracy values (MAE; RMSE; RAE; R2) compared to existing models in each china dataset (923.330; 1470.536; 0.436; 0.848) and desharnais dataset (9.540; 11.694; 0.041; 0.978), respectively. The proposed model generates synthetic data which is similarity to real data in the software effort estimation context. In addition, synthetic data improves the quality of performance on baseline machine learning in software effort estimation contexts rather than using real data.

398

Authors: Seema Sabharwal, Priti Singla

Sign language translation through deep learning is a popular topic among researchers nowadays. It opens the doors of communication for deaf and mute people by translating sign language gestures. The translation of input sign language gestures into text is called the sign language translation system (SLTS). In this paper, optimised machine learning-based SLTS for Indian sign language (ISL) has been proposed to facilitate deaf-mute persons. Further, this paper presents a simulation analysis of the impact of the number of convolution layers, size of stride function, epochs, and activation function on the accuracy of translation of ISL gestures. An optimised ISL translation system (ISLTS) for fingerspelled alphanumeric data of 36 classes using a convolution neural network (CNN) with a novel RADAM_NORM optimiser has been proposed. The proposed system has been implemented using two datasets- the first customised ISL alphanumeric dataset has been taken from Kaggle and the second dataset has been prepared by the author consisting of 36 classes and nearly 50K images. The accuracy of the proposed ISLTS on the first dataset is 99.446 % and on the second dataset is 97.889%.

422

Authors: Ibrahim A. Zedan, Mona M. Soliman, Khaled M. Elsayed, Hoda M. Onsi

SIFT-based techniques have achieved satisfying performance in detecting copy-move forgery (CMF). Typically, these techniques find the matched regions of an image and re-examine them using a variety of methods to determine whether CMF has occurred or not. However, these techniques have some shortcomings related to how they handle false matches, which usually occur due to image continuity or self-similarity. First, a spatial distance threshold or segmentation-based methods are commonly utilized to handle image continuity. Second, several external methods along with manually created thresholds are utilized to handle image self-similarity. In this paper, we propose a new matching strategy that is resistant to false matches while reducing reliance on external methods and thus avoiding several thresholds. We model the keypoint as a whole region rather than a single point and employ the intersection over union measure to deal with image continuity. To reduce false matches caused by image self-similarity, we combine the cross-matching test with a modified distance ratio test. This combination takes into account the ability to detect multiple cloning. Moreover, we utilize a support vector machine to learn the threshold(s) needed to decide if CMF has occurred or not. The proposed methodology is evaluated over three challenging datasets: MICC-F600, Coverage, and MICC-F220. On MICC-F600 dataset, our proposed methodology outperforms other state-of-art techniques and achieves high precision of 99.38%, recall of 97.5%, and 98.42% of F1 score. Additionally, the comparative evaluation using Coverage, and MICC-F220 datasets proved the effectiveness of the proposed methodology to handle a variety of attacks.

450

Authors: Suadad Zaidan Khalaf, Mohamed Ibrahim Shujaa, Ahmed Bahaaulddin A. Alwahhab

Law enforcement and civilian protection are increasingly dependent on surveillance technologies. To prevent social, economic, and environmental harm, video surveillance situations like those in train stations, schools, and hospitals must automatically detect aggressive and suspicious behavior. Intelligent video surveillance systems now enable human interaction monitoring. Despite its security benefits, crowds and the camera's vision make it difficult to distinguish regular from abnormal activity. Therefore, there is a great deal of investigation into the many methods of identifying violent behavior. The detection approach presented in this research can be broken down into three distinct subfields. The classification methods employed led to the formation of these classes. Classifications include Gradient Descent (SGD), Adaptive boosting (ADA), Naive Base (NB), and Logistic Regression (LR) for detecting violent acts with machine learning. Methods for feature extraction and violence detection are derived using Principal Component Analysis (PCA). Dataset learning- based models for violence detection can also be evaluated using the AIRTLab dataset, a model developed to test the robustness of algorithms against false positives. In order to determine if the proposed model's resistance to false positives, accuracy (ACC) metrics were computed on the proposed model and used to set a benchmark for the AIRTLab dataset. According to the research literature, the proposed models are accurate of 98.31% with the logistic regression classifier having superior generalization skills.

472

Authors: Lusia Rakhmawati, Hapsari Peni Agustin Tjahyaningtijas, Wiyli Yustanti, Wiryanto Wiryanto

Due to extensive Internet usage, digital media can be quickly altered, copied, and distributed via information networks. As a result, digital information's illicit reproduction became a serious issue. As a means of copyright protection for digital media, digital watermarking has been demonstrated to be effective. Based on a discrete cosine transform (DCT), an improved robust watermarking technique considering image characteristics, a 1-D linear transformation, and an optimal embeddable AC coefficient is proposed in this paper. The host image is first divided into 8 x 8 non-overlapping block and block selection for DCT transformation is performed. The variance in each block is used to identify the embedding watermark location. A linear map is used to scramble the watermark image, which is embedded into the AC coefficients which has a maximum difference between embeddable and non-embeddable non-zero AC coefficients. Experiment results show that the suggested approach can withstand a range of a non-malicious attack with very satisfactory results for BER close to 0, and NC near to 1.

497

Authors: Noora Nazar, Qusay Al-Doori, Omar Alani

Low density parity check (LDPC) codes are highly regarded for their exceptional error correction capabilities, making them a preferred choice for error correction coding (ECC) in modern communication standards. These codes are implemented in specialized integrated circuits (ICs) that facilitate data transmission. However, the power requirements of complementary metal-oxide-semiconductor (CMOS) circuits, which operate across a wide frequency range, present a significant challenge for future communication systems. To address this challenge, this paper proposes a novel multi-frequency power reduction strategy for LDPC encoders. The strategy utilizes dynamic voltage and frequency scaling (DVFS), a well-established power reduction method in CMOS circuits. By combining DVFS with fuzzy logic control, the system optimizes the voltage supplied to the LDPC encoder, achieving substantial power reduction while maintaining coding efficiency, flexibility, and performance. The proposed approach dynamically adjusts the encoder's voltage levels based on real-time conditions, ensuring efficient power consumption across different frequencies. This innovative strategy aims to overcome the power challenges faced by future communication generations, which have been largely overlooked in previous research efforts. This paper aims to assess the effectiveness of a multi-frequency power reduction strategy for LDPC encoders. Through extensive analysis and evaluations, it demonstrates significant power reduction while maintaining LDPC coding efficiency. These findings contribute to the development of power-efficient LDPC encoder designs, meeting the evolving needs of communication systems. Results reveal the system's performance across different frequency ranges. In the low-frequency range (1 MHz to 100 MHz), a remarkable 90% power reduction is achieved. In the medium-frequency range (100 MHz to 1 GHz), the reduction is 50%, while in the high-frequency range (1 GHz to 5 GHz), it reaches 20% to 14.5%. Notably, the system operates most effectively in the low and medium frequency range, providing substantial power savings. These results highlight the potential of the multi-frequency power reduction strategy for LDPC encoders in addressing power challenges across different frequencies. The significant power reduction achieved demonstrates its effectiveness and adaptability to varying frequency demands. This contributes to the development of power-efficient LDPC encoder designs, optimizing performance and energy consumption in various communication systems.

523

Authors: Dhiza Wahyu Firmansyah, Riyanarto Sarno

Breath acetone concentrations were found to be correlated with blood ketone levels. Based on this evidence, predicting blood ketone levels using breath analysis and machine learning (ML) becomes possible. Nevertheless, a good ML model requires a large amount of training data. Under certain conditions, it is difficult to collect large amounts of data such as during the Covid-19 pandemic. To overcome this problem, we propose an augmentation technique to extend the number of training datasets using two step synthetic minority oversampling (SMOTE). The first step was to increase the amount of training data by combining it with synthetic data, while the second step was to balance the data at each ketone level. The strategy for using SMOTE with regression was further explained since this study aims to predict ketone levels with numerical output values and SMOTE is typically used in classification cases. The proposed method was evaluated by entering the data into several ML methods such as deep neural network regression (DNN-R), linear regression (ML-R), ransac regression (RC-R), K-nearest neighbour regression (KNN-R), decision tree regression (DT-R), random forest regression (RF-R), Ada boost regression (AD-R), Gradient boost regression (GB-R) and XG-boost regression (XGB-R). Based on the test results, when compared without the proposed method, an increase in accuracy was obtained on DNN-R, ML-R, RC-R, KNN-R, DT-R, RF-R, AB-R, GB-R, and XGB-R by 0.958%, 9.51%, 35.74%, 18.133%, 8.236%, 11.348, 9.47%, 5.093%, and 11.264% respectively.

548

Authors: Wissal Farsal, Mohammed Ramdani, Samir Anter

The graded multi-label classification (GMLC) is an extension of multi-label classification. Whilst a multi-label classifier is limited to predicting the set of relevant labels, a graded multi-label classifier predicts the degree of relevance of a set of given labels. A key challenge of this learning problem consists of modelling the dependencies among the labels to improve the predictive accuracy. The algorithm adaptation-based solutions, which modify the algorithms directly to handle GMLC were proven effective in modeling these dependencies in comparison to the transformation-based models which reduce the graded multi-label datasets into a set of multi-class or binary datasets. In this paper, we propose an adaptation of random forest algorithm (GML_DT) with an adapted CART (classification and regression trees). The adapted algorithm is based on a modified formula for the Gini Index which fully models the label dependencies. The performance of the new model was tested on a 101 benchmark datasets and compared against the most influential methods for GMLC. The evaluation metrics considered in the experimental study are proper to the graded multi-label setting, i.e., hamming loss and vertical 0-1 loss. The experimental results show that the proposed model outperforms the considered models for the graded multi-label performance metrics. The increase in performance and overall accuracy is quantified by a decrease in the hamming loss of more than 13% and a decrease in vertical 0-1 loss that exceeds 10% when compared to the state-of-the-art models.

568

Authors: Christy Atika Sari, Muhammad Hafizh Dzaki, Eko Hari Rachmawanto, Rabea Raad Ali, Mohamed Doheir

The exchange of confidential information should be done in a secure environment. Therefore, security is needed if the exchange of information is carried out using Internet media. The way to secure the information can be done using cryptography and steganography techniques. and vigenere cipher had been applied as cryptography techniques and the LSB method as steganography techniques. This research has also compared the effect of the Fibonacci sequence in the encryption process. In this study, six images were used, each with a different extension, size, and type of images. All these images will be inserted with the secret message resulting from the encryption that applies the Fibonacci sequence or without applying the Fibonacci sequence. Fibonacci series is used to generate a random key, which is used for encrypting the message in playfair encryption algorithm. The level of effectiveness of the Fibonacci sequence has been determined and measured using MSE, PSNR, Entropy, UACI, and NPCR. The highest PSNR was in a 512x512 pixel grayscale image with a result of 7.9632, the best PSNR obtained was more than 85 dB, while the best UACI obtained was more than 49% using Fibonacci and 34% without implementing Fibonacci. The NPCR obtained is both above 99%, it is just that the image that has been implemented by Fibonacci produces a slightly higher UACI.

591

Authors: Balaji Sreenivasulu, Anjaneyulu Pasala, Gaikwad Vasanth

Early discovery and precise classification of plant-lead diseases are used to handle their spread and enhance the overall yield and product quality. The deep learning approach attains significant achievements in image classification and recognition. Since the classification using deep learning mainly depend on large-scale dataset for preventing the overfitting issue. Image augmentation is required to be developed to eliminate the risk of overfitting during the classification. In this research, the deep learning model based augmentation namely uncertainty based progressive conditional generative adversarial network (UPC-GAN) is developed for improving the plant leaf disease classification. The UPCGAN is used to map the images from one domain to another domain in a paired manner and estimate the uncertainty with the created images. Moreover, UPCGAN performs pixel wise residual distribution using the independent distributed zero mean generalized Gaussian distribution (GGD). The progressive learning of UPC-GAN increases the differences in the augmented synthetic images for improving the classification using DenseNet121. The dataset used to evaluate the proposed UPCGAN-DenseNet121 method is PlantVillage dataset. The performance of UPCGAN-DenseNet121 is analysed using accuracy, precision, recall and F1-score. Existing research such as deep convolutional GAN (DCGAN)-GoogleNet, conditional GAN (CGAN)-DenseNet121 and Fast wide and deep feature extraction block (WDBlock) based GAN namely FWDGAN are used to evaluate the UPCGAN-DenseNet121. The accuracy of UPCGAN-DenseNet121 for 10 classes is 98.2%, which is high when compared to the DCGAN-GoogleNet, CGAN-DenseNet121 and FWDGAN.

611

Authors: Fatima Abdul Sattar AL-Taie, Ahmed Sabah Al-Araji

In this research, a predictive voltage neural controller and remote monitoring for the nonlinear proton exchange membrane fuel cell (PEMFC) system are implemented in real-time. This paper's major purpose is to precisely and rapidly determine the appropriate hydrogen partial pressure (PH2) control action to improve the fuel cell's nonlinear performance under varying load currents and to enable remote monitoring of the nonlinear PEMFC system based on the internet of things (IoT). The laboratory, virtual instrument engineering workbench (LabVIEW) package is used to demonstrate the real-time performance of the proposed predictive voltage controller applied to the 150-watt PROTIUM PEMFC, which will be used to generate the appropriate amount of hydrogen partial pressure control action that will enter the fuel cell for stabilizing the desired output voltage. The message queuing telemetry transport (MQTT) protocol and a Raspberry Pi 4 acting as a local server are the building blocks upon which the monitoring component of the proposed system is implemented in order to monitor the desired output voltage, the fuel cell output voltage, and PH2. The Raspberry Pi collects the necessary fuel cell data and sends it to the node-red dashboard for monitoring. According to the simulation and the experimental results obtained using the proposed predictive controller on PROTIUM PEMFC, the proposed controller can generate an accurate, prompt, and timely reaction to the hydrogen partial pressure control action to reduce the tracking voltage error and to get rid of the fuel cell output voltage oscillation. The proposed experimental work was compared to the simulation findings to confirm its effectiveness in terms of effectively tracking the desired output voltage, providing a fast response, and achieving the optimal partial pressure of hydrogen. However, in the simulation findings, a voltage error of 0.01 volts was observed without any oscillation. On the other hand, the experimental results indicate a slightly higher voltage error of approximately 0.1 volts, accompanied by oscillations of around ± 0.1 volts.

642

Authors: Yamamh Alaa Alamari, Ahmed Fanfakh, Esraa Hadi

Currently, there are around 4.95 billion people who use the internet, which creates a large audience with an increasing demand for activities that can be done online. Some examples of these activities include social networking, information sharing, and online shopping. Therefore, there is an urgent need for improved levels of secrecy and privacy. It has recently come to light that one of the most serious challenges to the mainstream adoption of business apps is that of online fraud. As a direct consequence of this, issues pertaining to authentic cation, authorization, and identification have emerged as critical considerations in today's open and accessible society. The act of recognizing an entity, whether it be a human, a computer, or a software program, is referred to as the identification process. Authentication and authorization are complementary processes that are used in security systems to decide which persons are permitted to access information resources over a network. Authentication is performed by the user's device, while authorization is performed by the security system. However, a variety of potential solutions have been suggested, one of which is the use of parallelism in order to boost the efficiency of authentication methods. This work proposes a new message authentication algorithm that is applied in parallel over a multicore processor. The proposed message authentication algorithm uses two PNGRs and two substitution boxes to encrypt and authenticate the plain message. The encryption process has a one-round operation, making it a fast technique to encrypt and decrypt blocks of messages. In comparison to the existing method, the proposed authentication method outperforms the parallel speck-based authentication method by an average of 3.27 times faster when executed over a multicore CPU. The average speedup compared to the sequential version of the proposed algorithm and its parallel implementation is 2.99. The proposed method passes the most difficult randomness test, and the obtained MAC values are tested further to meet other security measurements.

669

Authors: Al_hussein M. Alturfi, Sonia Goyal, Amrit Kaur

Artificial bee colony (ABC) is a very popular algorithm and has been widely used in the literature. The algorithm has some inherent drawbacks, including slow convergence, and poor exploration versus exploitation, among others. In order to deal with these problems, we are hybridizing the ABC algorithm with the JAYA algorithm. The proposed algorithm has the added properties of both ABC and JAYA and has been named as JABC algorithm. The key idea is to add prospective equations of JAYA into ABC for better exploitation, and adding new mutation operators for enhanced exploration and better convergence. The proposed algorithm is tested on CEC 2005 benchmark problems and real world synthesis of linear antenna array (LAA). SLL reduction in five different LAA’s is done, for position, amplitude, and phase optimization. We use 10-element, 24-element, 28 element and 40-element LLA for optimization. For performance evaluation, the algorithm is tested with respect to basic ABC, JAYA, spider monkey optimization (SMO), Moth flame optimization (MFO), Chameleon Swarm Algorithm (CSA) and other algorithms. Experimental results on both benchmarks and LAA problems, show that JABC is significantly better as compared to others. Also, statistical analysis using Wilcoxon’s test and Friedman tests shows the superior performance of JABC algorithm with respect to other algorithms.