International Journal of Intelligent Engineering and Systems

1

Authors: Salsabila Basalamah, Agus Sihabuddin

Outliers can cause the results of the analysis to be biased. Two approaches to dealing with existing outliers are removing the outliers or modifying the method used. Commonly used methods like machine learning (ML) often require enhanced robustness in predicting outliers. One such method is decision tree regression (DTR). However, the DTR method has limitations as it does not consider outliers and makes predictions at leaf nodes based on central values of the data, which can introduce biases into the results. One of the algorithm that retains outliers is the M-estimator from robust regression. This study proposes a modification of the M-estimator for DTR by using Huber weights on leaf nodes for DTR predictions. We used five regression datasets sourced from UCI. The results are that the dataset with outliers provides better predictions on the concrete dataset, superconductivity dataset, Boston dataset, and Airfoil dataset having the best mean absolute error (MAE) of 3.963, 9.140, 2.021, and 1.644, with QSAR fish toxicity the only exception, where has the best MAE of 0.522 for the outlier remover dataset.

20

Authors: Susmitha Alamuru, Sanjay Jain

In recent years, video event prediction and classification is considered a hot research topic among researchers, due to its social influence and its extensive real-time applications. In this research, a video event classification system is proposed to predict and categorize various day-to-day events. This research proposed an improved pelican optimization algorithm (IPOA) to optimize the hyper-parameters which aids in better classification accuracy. The input data is gathered from three well-known datasets such as University of Central Florida101 (UCF101), Human Motion Database 51 (HMDB51) and columbia consumer video (CCV) dataset. The raw data is pre-processed using the data normalization technique and the DenseNet 201 model is used to extract the features from the pre-processed output. The extracted features are fed into the stage of feature selection using the improved grey wolf optimization (IGWO) algorithm. The hyper-parameters of the selected features are optimized using the proposed IPOA algorithm and classification takes place using bidirectional long short term memory (Bi-LSTM) classifier. The experimental results show that the proposed IPOA with Bi-LSTM achieved a better classification accuracy of 93.91%, 94.19%, and 90.19% for datasets such as UCF101, HMDB51 and CCV respectively which is comparatively higher than the existing techniques such as improved residual convolutional neural network (CNN), Bi-LSTM CNN, two-stream 3D CNN model and gait event detection system.

41

Authors: Nejood Hashim Al-walidi, Nagy Ramadan Darwish

Traceability link recovery is a crucial task in software engineering that ensures the development of dependable and credible software systems. Traceability links between requirements and source code support various activities in the software development process, including change management and software maintenance. These links can be established manually or automatically. Manual trace retrieval is a time-consuming task. Automatic trace retrieval can be performed via various tools such as information retrieval or machine learning methods. Some automatic tools couldn't retrieve the links between requirements and source code. Meanwhile, a big concern associated with automated trace retrieval is the low precision problem primarily caused by the term mismatches across documents to be traced. This study proposes an approach that addresses the low precision problem caused by the term mismatch problem between requirements and source code to obtain the greatest improvements in trace retrieval accuracy. The proposed approach utilizes a variational autoencoder (VAE), an unsupervised deep-learning model in the automated trace retrieval process. We have conducted a series of experiments on three datasets: eTour, SMOS, and eANCI to evaluate our approach against existing approaches. In order to validate the effectiveness of our proposed approach, we compared it to three previous studies that addressed the same problem and utilized the same datasets: the first study used unsupervised machine learning based on clustering, the second study used active learning, and the third study used a classification machine learning. The results show that our proposed approach improves the trace retrieval precision in the automated trace retrieval process.

65

Authors: Raya K. Mohammed, Nasser N. Khamiss

The orthogonal multiple access (OMA) used in fifth generation (5G) is suffered from limited resources. Due to the increasing number of users and ultra-high data rate requirements for user applications, non-orthogonal multiple access (NOMA) is a promising scheme for future communication systems. NOMA suffered from inter-user interference due to non-orthogonal resource sharing. This paper proposes the enhanced NOMA technique to be employed in the 5G system to satisfy the data rate requirements for Beyond 5G (B5G). The performance improvement is achieved through incorporating the filtering techniques with the new-radio orthogonal frequency division multiplexing (NR-OFDM) in 5G used waveform. The limitation in high out of band emission (OOBE) in the existing waveforms can also be reduced through the proposed method. Performance evaluation of the proposed system is demonstrated, and the results proved an improvement in bit error rate (BER) and power spectral density (PSD). As compared with the literature, the proposed system provides almost the same error floor with a benefit of approximately 11dB and 17dB SNR for full-band and sub-band-based filtering techniques, respectively. The achieved side lobe gain is up to 257dB for the Kaiser-Chebyshev-based filtering method.

83

Authors: Deena Sivakumar, S. Suganthi Devi, T. Nalini

Wireless sensor networks (WSNs) can be defined as a set of independent sensor devices, which are linked by wireless channels. WSN has potential for real-time monitoring namely environment surveillance, military applications, industrial applications, health monitoring, and many more. Certain limitations of WSNs include limited storage space and node energy. Energy efficiency will be the major design problem in WSN, which is solved through clustering and routing methods. They were regarded non-deterministic polynomial (NP)-hard optimizing issues and can be solved with the utilization of metaheuristic techniques for identifying the near-optimal or optimum solutions. This article introduces a novel yellow saddle goatfish optimization algorithm based energy aware multihop routing (YSGOA-EAMHR) method for WSN. The presented YSGOA-EAMHR methodology concentrates on the recognition of optimum routes to destinations for data communication in WSN. To attain this, the proposed YSGOA-EAMHR methodology is mainly based on YSGOA, which is stimulated by the characteristics of yellow saddle goatfish. In addition, the presented YSGOA-EAMHR model derives a fitness function involving residual energy (RE) and distance to BS. The YSGOA-EAMHR model selects the nodes with higher RE and lowers the distance to BS as optimal relay nodes for data transmission. The investigational validation of the YSGOA-EAMHR technique is examined utilizing a sequence of measures. The comparison study reported that the YSGOA-EAMHR technique has gained significant performance over existing models such as FUCHAR, GWO, MOPSO, WGWO, and HMBCR techniques.

108

Authors: Arunachalaprabu Gurunathan, Fathima Bibi Kaja Moideen

Biological data analyses involve researchers from several fields. To store and manipulate the huge volume of biological data obtained from different aspects is difficult. Compression algorithms considerably increase the storage medium's capacity while reducing the number of bits required representing the sequence. The core concept behind EIBDNASCA involves creating an optimized index file, which stores the non-repetitive bases (run length less than 8). This index file plays a crucial role in swiftly retrieving and reconstructing specific segments of the DNA sequence during the decompression process. In addition to the index file, EIBDNASCA incorporates a work file, which stores the repetitive bases (run length above 8) and represented in binary form. This work file allows the algorithm to perform various pre-processing and transformation tasks on the DNA sequence before generating the final compressed output. Finally, an enhanced Huffman coding technique is applied to the symbols present in the index file, optimizing the encoding process for more efficient compression. The proposed algorithm is examined using a variety of different GenBank database sources. Compression ratio, compression gain, and time required to compress and decompress the sequences are the metrics used to assess the performance. The experimental findings indicate that EIBDNASCA attains an average compression ratio of 1.23 bpb (bits per base) with an average compression gain of 84.52%. The average compression time is recorded at 0.590 seconds, and decompression is completed in 0.625 seconds.

129

Authors: Sukesh Hoskote Aswathanarayana, Sundeep Kumar Kanipakapatnam

Melanoma skin cancer is the most life-threatening and fatal disease in the family of skin cancer diseases. Detecting melanoma at its early stage can improvise survival which requires an effective classification technique to categorize dermoscopic images as melanoma and non-melanoma. Based on the quality of the extracted features, the classification accuracy of the classifier is determined. However, the classification accuracy of the existing approaches is poor due to the presence of improper image boundaries and low quality. To overcome the fore-mentioned issues, this research introduced an optimization-based feature selection approach using the modified golden jackal optimization (MGJO) algorithm. The pre-processed image is segmented using a semantic mathematical model known as saliency-based level set with improved boundary indicator function (SLSIBIF) and the feature extraction is performed using the GoogleNet architecture. After this, the proposed MGJO algorithm was used to select the relevant features which aid in precise classification performed using multiclass-support vector machine (MSVM). The obtained results show that the proposed MGJO-MSVM achieves enhanced classification accuracy of 98.89 % for the ISIC-2017 dataset whereas the accuracy of the existing feature adaptive transformer network (FAT-NET), multi-attention fusion convolutional neural network-based skin cancer diagnosis (MAFCNN-SCD), W-net inception residual network, region-based convolutional neural network with fuzzy k-means clustering (RCNN-FKM) and gated fusion attention network is 93.26%, 92.22%, 96.97%, 95.6% and 93.97% respectively.

149

Authors: Abhishek Dwivedi, Nikhat Raza Khan

This research paper aims to forecast equipment's remaining useful life (RUL) to improve maintenance planning and reduce costs. This paper presents the spectral-based transformer (SPT) model, designed for predicting the remaining useful life (RUL) in the evolving maintenance landscape of industry 4.0. Proactive maintenance is becoming increasingly important as it improves performance and reduces losses. SPT utilizes advanced attention mechanisms and innovations, which have been evaluated on the C-MAPSS dataset to simulate various operations. The contributions include discrete cosine transform attention (DCTA), uninhibited positional and contextual attention (UPCA), multi-head shortcuts, and bidirectional structures. Component efficacy is rigorously assessed through ablations. The results demonstrate that SPT exhibits superior performance compared to other methods, with a notable advantage on the challenging FD002 and FD004 sub-datasets within the C-MAPSS dataset. The proposed method decreases the root mean square error (RMSE) by 14% and enhances the performance scores of FD002 and FD004 by 10% and 24%, respectively. Additionally, it reduces the RMSE of FD004 by 15%. The model outperforms current methods, showing stability and generalization across different subsets of data. SPT demonstrates proficiency in capturing degradation patterns, which shows the potential for accurate remaining useful life (RUL) prediction. This tool is designed for time-series regression and has potential applications in various industries. Future research could focus on expanding the system's capabilities to process higher frequencies and broader contexts effectively. The SPT method provides a thorough approach for predicting the remaining useful life (RUL). It can potentially improve maintenance decisions and system performance in the context of Industry 4.0.

169

Authors: Mohammed H. Ahmed, Asaad N. Hashim, Khalid A. Hussein

Cryptography plays a crucial role in safeguarding privacy and confidentiality. Therefore, it is imperative to ensure the security and resistance to cryptanalysis of the employed ciphers. In this paper, an innovative approach called three levels multiple particle swarm optimization (TL-MPSO) has been proposed to enhance the traditional particle swarm optimization (PSO) for cryptanalysis of nonlinear stream ciphers through a ciphertext only attack. We focus on the geffe generator system, which relies on a set of linear feedback shift registers (LFSRs) with a nonlinear combining function. TL-MPSO employs a three-levels approach with distinct search strategies. In the first level, known as the discovery level, particles are randomly distributed into equally sized sub-swarms with no communication between particles in different sub-swarms. The primary objective is to thoroughly explore the search space, with communication limited to neighbouring particles within the same sub-swarm. The second level, called the extraction level, is generated artificially by perturbing the best particles from the discovery level. Particles in different sub-swarms communicate in parallel and co-evolve. Finally, the third level, the optimization level, is artificially generated from the previous level, with all sub-swarms merged into a single large swarm. In all levels, an adaptive movement strategy for each particle is established through a new inertia weight updating process, and a dynamic regrouping strategy is applied to enhance the chances of achieving the global best. Evaluation results demonstrate that the proposed system requires only a minimal number of characters and achieves optimal CPU times. Specifically, when attacking Geffe generators with lengths ranging from 15 to 35, the proposed algorithm consistently achieves optimal solutions with just 10 to 50 cipher characters. Our experiments reveal that TL-MPSO is more accurate and faster than traditional PSO, representing an efficient technique compared to existing cryptanalysis methods.

193

Authors: S. Priya, R. S. Ponmagal

As cloud computing is a combination of variety of technologies, there are a number of security concerns need to be addressed. To protect from various attacks and to improve the security of cloud users, it is required to analyze the trust relationships among the cloud resources and tasks. Most of the trust management systems did not consider the significance of interactions, which affects the correctness of trust evaluations. This paper proposes Gaussian flower pollination optimization (GFPO) based trusted service selection (TSS) technique. In this technique, each cloud service provider’s reputation values are computed from the trust value of its servers. The individual trust values of cloud servers are determined based on interaction success rate (ISR), service success index (SSI) and service response time (SRT) parameters. The optimal weight values of these parameters are adaptively determined using GFPO algorithm. During the data transfer from server to the user, the user checks the server’s trust value and ignores the response if its trust value is low. Experimental results from Cloudsim show that GFPO-TSS minimizes the computation overhead and access delay, maximizes the accuracy and success rate. GFPO-TSS has 45% reduced computation cost, 10% increased data correctness.

212

Authors: Irwan Budi Santoso, Shoffin Nahwa Utama, Supriyono

Detection of brain tumors based on magnetic resonance imaging (MRI) images is essential for follow-up examinations. Several CNN models have been proposed before to get the best performance in detecting brain tumors. However, it is still necessary to improve performance due to complex brain structures, varying tumor shapes and sizes, and the position of brain tumors. Therefore, we propose a new voting of convolutional neural networks (CNNs) based on MRI images to detect brain tumors. CNN in the proposed method has three network paths, each involving a convolution process with a different kernel size. Involving different MRI image input shapes on the proposed CNN can provide different detection results, so voting is needed for the final detection. We propose a voting method with the condition that if only one proposed CNN model with a particular input shape detects a brain tumor in the MRI image, the final detection result indicates a tumor in the image. To evaluate the method's performance, we used brain MRI image datasets for tumor detection arranged into training, testing 1 (small size), and testing 2 (large size). The test results on the dataset show that the proposed method yields the best accuracy of 99.24% for testing 1 and 99.92% for testing 2. With these results, our proposed method performed better than the other methods, including VGG16, VGG19, ResNet50, MobileNetV2, InceptionV3, and Xception.

240

Authors: Suhas Shankarnahalli Krishnegowda, Hosanna Princye Periapandi

High-efficiency video coding (HEVC) stands as one of the most extensively employed video coding standards, with the primary encoding challenge revolving around the searching process. To minimize coding complexity of HEVC block partition, a new algorithm is proposed for supporting the partition architecture of coding tree unit (CTU) in intra-coding. In this research, problem of partition architecture decisions of CTU is resolved through two stages. In first phase, the bagged tree method is evaluated for predicting CTU splitting, and in the second phase, the partition issue of 32×32 size CU is given as 17 outcome classification. To attain a huge accuracy of prediction, a residual attention-based long short-term memory (LSTM) with parametric rectified linear unit (AT-LSTM-PReLU) is proposed. The proposed method produces a structure of partition quad-tree of CTU that makes multiple decisions at variant depth levels. The performance of proposed method is estimated with a dataset gathered from the joint collaborative team and video coding (JCT-VC). The proposed approach achieved a remarkable bit rate improvement of 3.72%, a significant ∆T enhancement of 97.28%, and minimized video quality degradation by just -0.05dB. These results outperform other deep learning methods like convolutional neural network (CNN), recurrent neural network (RNN), and long short-term memory (LSTM).

263

Authors: O. Homa Kesav, Rajini G. K

In this study, we propose a novel enhanced deep learning method for the detection and classification of brain tumours known as the reduced complexity spatial fusion CNN (RCSF-CNN) method. This approach integrates complexity feature extraction, which improves the quality of feature extraction from brain tumour pictures. To capture crucial detection properties, image variables such as mean, standard deviation, entropy, variance, smoothness, energy, contrast, and correlation are extracted. These attributes are then employed by the RCSF-CNN to detect and categorise brain cancers. When paired with the discrete orthogonal stockwell transform (DOST) as an intermediary stage, the suggested method illustrates the effectiveness and superiority of the augmented deep learning methodology for brain cancer identification. The studies were carried out using the BRATS dataset via Kaggle, with the network trained on 32 samples and the features of five sample pictures assessed. The RCSF-CNN stands out for its efficient architecture, which includes spatial fusion as well as a critical normalisation step. The addition of class activation mapping (CAM) increases transparency and interpretability, highlighting the model's innovation. The MATLAB simulation tool was used for implementation, and the experimental investigations were carried out on the free-source brain tumor image segmentation benchmark (BRATS) dataset. The results obtained in brain tumour identification reveal an entropy value of 0.008, an energy value of 0.8155, and a contrast value of 0.354. These entropy, contrast, and energy values are critical in the detection of brain tumors. Furthermore, in terms of accuracy, specificity, and sensitivity, the new technique beats earlier methods such as conventional CNN, deep learning with modified local binary patterns, and ML algorithms such as SVM in brain Tumour detection. The achieved accuracy of 98.99% indicates a high level of total correct classifications. The specificity of 99.76% illustrates the methodology's capacity to correctly identify non-tumor regions, while the sensitivity of 98.43% demonstrates its ability to correctly detect cancer locations.

288

Authors: Abbas Ahmad S M K M, Devanna H, Dustakar Surendra Rao, Mohammed Ali Sohail

Cognitive radio technology is introduced to identify the spectrum holes dynamically. Such dynamic spectrum access operations cause the nodes in the cognitive radio sensor networks (CRSNs) into energy depletion problem. To handle this problem or to utilize the entire network’s energy efficiently, clustering is found as one of the best solution. Even though uniform clustering mechanism reduce the energy consumption but it is not suitable when the network node density increases. Hence, this paper proposes a new clustering mechanism called as energy and spectrum aware clustering routing protocol (ESCRP) for CRSNs. Initially, the entire network is clustered non-uniformly into several clusters. Next, energy, channel availability rate, geographical and temporal correlation metrics are used to select the cluster head (CH) for each cluster. Finally, the collected information from each CM is transferred to the sink node through multi-hop communication between CHs. Extensive simulation experiments are carried out over the proposed ESCRP and the performance is measured with several performance metrics including Network lifetime and average energy consumed. From the experimental results, we observed that the weight combination α=1/8,β=1/8,γ=3/4 consumed less energy than the remaining combination weight values. Finally, the average energy consumed for the combination α=1/8,β=1/8,γ=3/4 for 2-layered network and 3-layered network is approximately 0.4J and 1.2J respectively.

316

Authors: Ruwaida Mohammed, Israa Shihab, Mustafa Musa

The digitization of goods and services is predicated on the production of value by contemporary organizations. What customers say about a firm on social media has become a significant part of the big data revolution. Furthermore, social media data analytics is a challenging topic to learn because of the partiality of text assessment and the extra variables. Social media data analytics using the deep learning (SMDA-DL) framework is proposed in this paper. A two-phase framework is suggested in the SMDA-DL framework: Pre-processing stage and selecting an efficient deep learning technique for this information is the emphasis of phase one; the second phase relies on existing components of large data structures and the model developed in phase one. First, small and large databases in non-big data sets will be analyzed in the first step of the suggested approach. While the first step of the deep learning model may be used to analyze massive data databases, the second stage is recommended. For the first layer of the architecture, there is a case study for assessing social media assessments. The proposed framework yields good simulation results in an accuracy ratio of 98.7%, precision ratio of 99.7%, recall ratio of 98.4% and F score of 98%.

343

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

Load balancing is critical to managing server resources efficiently and ensuring optimal performance in distributed systems. The weighted round robin (WRR) algorithm is commonly used to allocate incoming requests among servers based on their assigned weights. However, static weights may not reflect the changing demands of servers, leading to imbalanced workloads. To address this issue, this study proposes a dynamic mechanism for assigning weights to servers in the WRR algorithm based on the data rate and incorporates the least connection approach for the best result. The dynamic mechanism considers each server's real-time data rate, representing its current load. Servers with higher data rates are assigned higher weights to attract a larger share of incoming requests, while those with lower data rates receive lower weights to manage their loads effectively. This dynamic weight assignment allows the algorithm to adapt to varying workloads and achieve better load balancing across servers. To further refine the distribution of requests, the least connection approach is employed to handle tie-breaking situations and for more fairness in distributing the loads. The proposed algorithm is a hybrid of data rate and the least connection, it is evaluated through simulations and real-world experiments. The results demonstrate its superiority in achieving improved load balance compared to other algorithms, such as round-robin RR and traditional static-weight WRR algorithms. By dynamically adjusting weights based on data rate and employing the least connection approach, the algorithm optimizes server resource usage, minimizes response times, and enhances overall system performance in distributed environments.

370

Authors: Imam Robandi, Muhammad Ruswandi Djalal, Mohammad Almas Prakasa

This study proposes the application of a single-band power system stabilizer type 1A (SB-PSS1A) to enhance the damping ratio in multimachine systems. The mayfly optimization algorithm (MOA) method is employed to optimize the performance of the SB-PSS1A. The South Sulawesi, Southeast Sulawesi, and West Sulawesi regions of the Sulselrabar system are covered by the proposed control plan. To evaluate the effectiveness of the proposed controller, frequency domain analysis and time domain simulation analysis with varying load changes are presented. The accuracy of the MOA method is compared to the swarm intelligence method based on particle swarm optimization (PSO) and firefly algorithm (FA). The analysis results indicate an improvement in system stability, as demonstrated by an increased damping ratio, reduced oscillation overshoot in generator speed and minimum rotor angle, and faster settling time. By implementing SB-PSS1A on 14 generators using the MOA optimization method, a damping ratio of 0.5868 was achieved.

393

Authors: K Bhavana, V Rajeswari, K Lalitha, J Vijayanand, Srinivasarao Thumati

The growing threat of global warming necessitates immediate demand for the adoption of sustainable methods in all sectors, including agriculture. Most agricultural practices require quality and reliable power sources to maximise productivity. In this regard, this work proposes the efficient integration of photovoltaic systems (PVs) with agriculture distribution feeders (ADF) with multiple objectives. The main goals of this study are to maximise efficiency, energy loss cost and reduce greenhouse gas (GHG) emissions while working within various operational and planning restrictions. The ideal position and size of PVs for a specified ADF must account for both continuous and discrete variables, which is why this study introduces hybrid sine cosine optimization (HSCO). HSCO is designed to achieve both efficient search characteristics and a quick convergence rate by embedding the linear search route features of the particle swarm optimisation (PSO) and adaptive convergence factor (ACF), respectively. HSCO's effectiveness of the HSCO is demonstrated by tackling the challenge of optimal PV integration in a real-time 28-bus ADF and IEEE 33-bus feeder for different situations. The 28-bus feeder reduces cost and loss by 45.5%, 47.518%, and 48.3% for light, normal, and heavy loads. However, 33-bus optimum PV systems cut losses by 64.94% compared to basic scenario and better than existing literature works. A comparative analysis of the computational aspects of the proposed HSCO with basic sine cosine optimization (SCO), particle swarm optimization (PSO), northern goshawk optimization (NGO) and artificial rabbits optimization (ARO) is also presented. Notably, with global optima and reduced computational time, HSCO outperformed than other algorithms. Furthermore, ADF was observed with a loss reduction, good efficiency and significant reduction in GHG emissions with appropriate PVs integration. These figures are encouraging in terms of adapting the proposed methodology for real-time applications.

418

Authors: B Ravindhar, B Suresh Kumar, B Mangu

This research paper addresses the limitations inherent in conventional volt-VAR control (VVC) mechanisms, such as on-load tap changers (OLTC) and voltage and capacitor banks (CBs), when applied to the implementation of efficient conservation voltage reduction (CVR) strategies within active distribution networks. In this paper, we present a cost-effective strategy that synergizes the rapid-response capabilities of PV smart inverters with traditional VVC mechanisms to enhance CVR operations. This study delves into the optimal coordination between photovoltaic smart inverters and conventional VVC systems for effective CVR operations. The investigation takes into account the implications of distribution network reconfiguration (DNR) on the overall operational costs, encompassing grid power procurement, emissions, losses, and switching expenses related to OLTC, CB, and remote controlled switches (RCS). To attain the optimal solution, we leverage the recent advancements of the white shark optimization (WSO) technique, meticulously validating its performance through extensive trials conducted on the 119-bus distribution system. The empirical results vividly showcase the substantial advantages of integrating VVC devices and smart inverters, particularly in scenarios characterized by fluctuating load conditions. Notably, the synergistic approach achieves a remarkable 6.51% reduction in energy consumption. Additionally, active and reactive power loss reductions of up to 33.67% and 28.55%, respectively, underscore the potency of the proposed strategy.

447

Authors: Mohamad Haizan Othman, Belinda Chong Chiew Meng, Nor Salwa Damanhuri, Mohd Ezane Aziz, Nor Azlan Othman

Fat extraction is a crucial aspect of diagnostic analysis in T1-weighted magnetic resonance imaging (MRI) images. However, the accuracy is affected by image inhomogeneity. Inhomogeneity refers to variations in signal intensity across an image, which can be caused by uneven magnetic fields or abnormal fluids in MRI image. This study uses fuzzy C-means (FCM) algorithm for fat region extraction. However, FCM is struggle with regions of similar intensity. The objective of this study is to propose a method for inhomogeneity correction using adaptive disk structure element morphological (ADSEM) approach. This rectifies the impact of inhomogeneity-induced intensity variations. The method is then integrated with FCM for fat extraction. This approach overcome FCM's intensity similarity limitation, enhancing fat extraction accuracy. Comparative assessments highlight the integrated ADSEM-FCM method's superiority over FCM. The quantitative assessment for proposed method in term of accuracy, recall, precision and F1 score is 0.9246, 0.9777, 0.7740, and 0.8526 respectively.

469

Authors: Pardeep Kumar, Ankit kumar

With growth in world's population over the last several decades, the provision of medical care has emerged as the component of human life that is most essential. A considerable number of people are losing their lives much too early due to heart disease on an annual basis. It is crucial to detect the illness at its earliest possible stage in order to lessen the chance of dying from heart disease. This should be done as soon as possible. The enormous amounts of information generated for diagnostic purposes have made it possible to construct complex learning-based models for the early, automated diagnosis of cardiac issues. This has been made possible as a result of the availability of this information. Because of this, there have been substantial gains made in the accuracy of diagnostics. As a consequence of this, there have been significant breakthroughs made in medical technology. The traditional approaches to machine learning are unable to generalize their conclusions to new datasets since these datasets were not part of the training set. As a direct result of this, the trained model's ability to provide accurate forecasts is being negatively impacted as a direct consequence of this. This research suggests a Deep learning based optimum gradient non-linear mapping of features in addition to learning through bagging and boosting-based ensemble learning in order to improve the accuracy of detecting five distinct types and binary classification of heart disease. considerable enhancement in accuracy from 8-10% and in binary average 2-3%" indicates that the use of deep learning approaches like ResNet-50 and CNNs has led to a significant improvement in the model's classification performance. The 8-10% enhancement might refer to a multi-class classification problem where the accuracy metric is generally lower due to the increased complexity of the task, while the 2-3% improvement could relate to a binary classification task where accuracies are usually higher, and even small improvements can be hard to achieve and thus are quite meaningful.

493

Authors: M Shanmugam, V M Viswanatha, K B Raja

In the domain of computer vision, face recognition systems play a pivotal role, particularly in the context of security and access control. This research paper introduces a novel approach to face recognition that leverages the fusion of compressed hybrid domain grades at a matching level. It takes into consideration two distinct compressed hybrid domain algorithms, referred to as algorithm one and algorithm two, with the aim of enhancing accuracy across ORL, YALE, and JAFFE face databases. In algorithm one, a combination of histogram intensities, discrete wavelet transform (DWT), and double density dual tree discrete wavelet transform (DDDTDWT) is employed for feature extraction. Notably, this algorithm also segments the binary bits of image pixel decimal values into most significant bits (MSB) and least significant bits (LSB). DWT is subsequently applied to the MSB, while the histogram of oriented gradients (HOG) is used for the LSB matrices. Meanwhile, Algorithm two explores a distinct approach by utilizing the GIST descriptor concept on the LL-sub band of DWT and HOG features, which are later combined through convolution. Moreover, for the neural network classifier, a GWO-Cuckoo optimized architecture is employed, ensuring that the extracted features are optimally utilized for precise face recognition. In addition, the minimum redundancy maximum relevance (MRMR) feature selection method is integrated into the feature extraction process. This further enhances the discriminative power of the selected features. The results achieved through this proposed technique are highly promising. Specifically, for the ORL, YALE, and JAFFE databases, accuracy rates of 97.71%, 100%, and 100% are achieved, respectively, surpassing the performance of existing methods.

525

Authors: Aruna Kumari Koppaka, Vadlamani Naga Lakshmi

In recent years, big data analysis has been one of the hot research topics which attracts more researchers to work on it. Big data is comprised of highly sensitive data which is needed to be secured and monitored effectively. However, problems related to privacy and security have been provoked due to ineffectiveness in storing the data and relying on third parties. To overcome the issues of storing the data in a cloud environment securely, this research introduced a hybrid cryptography algorithm that is comprised with advanced encryption standard (AES), triple data encryption standard (TDES), and blow fish algorithm (BFA). Initially, the data from the user are acquired and uploaded to the cloud storage space, then a key is generated which helps in encrypting and decrypting the user’s data. Process of encryption and decryption takes place using the proposed hybrid cryptography algorithm which effectively securely stores the data and helps the user based on their needs. Moreover, the rivest, shamir, adleman (RSA) algorithm is used in the process of key generation which helps to encrypt and decrypt the messages. The efficiency of the proposed approach is compared with the existing triple data encryption standard (TDES), blow fish algorithm – spotted hyena optimization (BFA-SHO) algorithm and secure data deletion and verification (SDVC) scheme on based on ciphertext policy-attribute based encryption (CP-ABE). The results obtained through analysis exhibits that suggested methodology took 183450 ms for executing the file size of 20 MB whereas the existing BFA-SHO took 248107 ms for executing the same file size of 20 MB.

555

Authors: Putri Taqwa Prasetyaningrum, Purwanto Purwanto, Adian Fatchur Rochim

One strategy for enhancing the promotion of marketing content involves the incorporation of gamification techniques. Existing research that examines the utilization of gamification through the classification approach often overlooks the issue of class imbalance within the dataset. Class imbalance occurs when the number of instances in the minority class differs significantly from those in the dominant class. This discrepancy can cause the classification method to perform worse because most classifiers work best when the class distribution in the dataset is balanced. Multiple studies underscore the significance of addressing class imbalances as a pivotal step toward enhancing the performance of classification algorithms. This study aims to demonstrate the impact of the data imbalance challenge and to identify the superior resampling method for integration into the machine learning process. The dataset employed pertains to users of a mobile banking application in Indonesia and encompasses variables such as gamification elements, demographic information, psychological factors, and customer engagement metrics. The chosen resampling technique for this study is SMOTE-Tomek. The various classification methods utilized include linear regression, K-NN, CART, random forest, SVM, stacking ensemble, and XGBoost. Among these methods, the SMOTE-Tomek resampling approach proves most effective when combined with the stacking ensemble classification. This combination yields a remarkable accuracy rate of 98.7% during 10-fold cross-validation, along with an impressive geometric mean score of 0.99. The thorough evaluation findings demonstrated that the approach presented in this study has a promising practical applicability and effectively identifies unbalanced elementary games.

584

Authors: Nur Nafi’iyah, Chastine Fatichah, Darlis Herumurti, Eha Renwi Astuti, Ramadhan Hardani Putra, Agus Subhan Akbar

Forensic gender identification based on teeth is needed to identify incomplete victims, only bones and teeth. Previous studies identified gender by manually or semi-automatically measuring mandibular parameters based on landmark points. Gender identification, especially for victims of mass disasters, requires accuracy, so it takes longer, especially if there are many parameters to be measured. In addition, the observer's manual or semi-automatic measurements may give different results. This study proposes a new automatic approach to generate ten mandibular landmark points from panoramic radiographic images for gender identification. We propose a step, namely determining the centroid point of the mandibular image and using linear regression to predict ten mandibular landmark points such as the two condyle, two gonion, four ramus, and two body of the mandible. This study obtained panoramic radiographic images from the Academic Dental Hospital, Universitas Airlangga. We calculated the distance between expert mandibular landmark points and the predicted results to evaluate performance. The prediction of the landmark with the smallest average distance is the lower point of the mandibular body 1 pixel. In comparison, the predicted landmark with the most extended average length is a gonion of 10 pixels.

612

Authors: Fahmida Hossain, A. K. M. Zakir Hossain, Muhammad Ibn Ibrahimy, Jamil Abedalrahim Jamil Alsayaydeh

This article presents a 6×6 MIMO array optimized for sub-6 GHz mobile and wireless applications. Utilizing a partial ground plane (PGP), the array incorporates a single-element half-circle-slotted monopole antenna to achieve a wideband response. The compact design, measuring 75 × 150 mm² and only 0.508 mm thick, is engineered to fit 6.6-inch display smartphones. Operating between 2.82 and 5.95 GHz, the array offers a -10 dB bandwidth and a VSWR of less than 2. It also delivers port-to-port isolation less than -20 dB and a peak realized gain of 5.3 dBi, with an antenna efficiency of 71%. Comprehensive far-field radiation patterns are observed across all 6 ports. The array maintains an envelope correlation coefficient (ECC) under 0.004 and a diversity gain (DG) above 9.98, demonstrating robust performance. Specific absorption rate (SAR) results at 3.5 GHz confirms compliance with MIMO standards, with values as low as 0.13 W/kg. This high-performance MIMO array is a strong contender for sub-6 GHz 5G applications.

634

Authors: Niranjan Muchandi, Rajashri Khanai, Mandakini Muchandi

Achieving higher accuracy of spectrum sensing (SS) becomes a challenge in cognitive radio adhoc network (CRAN) due to presence of varied power levels, interference, multipath fading and hidden node problems. Cooperative sensing based SS address these challenges by making spectrum decision based on distributed sensing measurements from multiple nodes. A major problem in cooperative SS is that, false sensing measurements sent by compromised node can impact the spectrum access probability and network services. These false measurements also affect the channel utilization calculation and disrupt the QoS based routing protocols for cognitive radio (CR) networks. The compromised node can also compromise the privacy of SU by leaking communicated messages. In this work, we provide a cross layer QoS assured routing system for CRAN that is secure as well as protects privacy. The proposed solution solves the problems of false sensing measurements using Kolmogorov Smirnov test on spectrum measurements in both spatially and temporal contexts. Attacks on spectrum measurements are detected and filtered using sequential probability test. Cross layer interactions are used in decision on source rate control, multipath selection and adaptive queing to provide guaranteed QoS. The proposed solution is able to increase the spectrum sensing accuracy by at least 7.6% in presence of sensing attacks and increase packet delivery ratio by at least 4% compared to existing works.

654

Authors: Sanjeev Mewalal Dwivedi, Sunil B. Wankhade

Social media has gradually become the primary news source for people in recent years, providing convenience but also leading to the spread of false information. With the rise of media-rich social media platforms, fake news has evolved from single-text to multimodal formats, prompting increased attention to multi-modal fake news detection. However, most existing methods rely on representation-level features that are closely tied to the dataset, resulting in insufficient modelling of semantic-level features and a limited ability to generalize to new data. To address this issue, we propose a semantically enhanced multimodal fake news detection method that utilizes pre-trained language models to capture implicit factual knowledge and explicitly extracts visual entities to better understand the deep semantics of multimodal news. We also extract visible features at different semantic levels, use a text-guided attention mechanism to model semantic interactions between text and images, and integrate multimodal features. Experimental results on real datasets based on Weibo news demonstrate that the proposed method outperforms other methods with an accuracy of 0.895 in multimodal fake news detection.

679

Authors: Mamta Tiwari, Hemant Kumar, Navin Prakash, Sunil Kumar, Rahul Neware, Shivneet Tripathi, Ritesh Agarwal

The tomato holds significant global significance as a crop, fulfilling crucial roles in agriculture and meeting essential human dietary requirements. The persistent presence of diseases poses a significant risk to tomato crops, compromising their productivity and economic viability. This study presents a novel approach that combines the vision transformer (ViT) with deep neural networks (DNN) to identify tomato leaf diseases. This research aims to develop a methodology for identifying tomato leaf diseases, specifically focusing on enhancing the model's performance and interpretability. The proposed technique integrates an improved multi-head self attention mechanism with L1-norm attention, enhancing its ability to capture essential image features. The capability of ViT to process images as sequences of patches is harnessed and integrated with the interpretative and classifying prowess of DNN. To train and test the model, we utilize a comprehensive dataset of 18,160 tomato leaf images classified into ten different disease types. According to the experimental findings, the proposed model achieved an exceptional accuracy rate of 99.74%. The proposed method surpasses other contemporary techniques in terms of both accuracy and precision (99.66%), recall (99.39%), and F1-score (99.53%) for disease classification. The model's exceptional performance highlights its potential for practical implementation in diagnosing tomato diseases. In conclusion, this study presents a novel and efficient method for detecting tomato leaf diseases, which fulfills the current agricultural need for scalable and precise techniques. The fusion of vision transformers and deep neural networks offers a robust solution with exceptional accuracy and interpretability, contributing to the sustainable cultivation of this crucial crop.

698

Authors: Purba Daru Kusuma, Meta Kallista

There has been a massive development of metaheuristic algorithms in the latest decade where swarm intelligence becomes the fundamental approach. Meanwhile, there is still no ideal metaheuristic that can solve all problems superiorly, as declared in the no-free-lunch (NFL) theory. This work introduces a novel swarm-based metaheuristic named as migration-crossover algorithm (MCA). In MCA, the swarm intelligence is enriched with the crossover technique and the neighbourhood search with unbalanced local search space. The global finest solution becomes the reference in the first step while the middle between two stochastically chosen solutions becomes the reference in the second step. The neighbourhood search is performed in the third step. The collection of 23 functions become the use case during the evaluation of MCA. In the first evaluation, MCA is compared with five new metaheuristics: total interaction algorithm (TIA), osprey optimization algorithm (OOA), migration algorithm (MA), coati optimization algorithm (COA), and walrus optimization algorithm (WaOA). The result reveals that MCA is finer than TIA, OOA, MA, COA, and WaOA in 20, 19, 17, 20, and 17 functions subsequently. The result of the second evaluation reveals that the global finest solution becomes the dominant contributor in the high dimension functions while the middle between two stochastically chosen solutions becomes the dominant contributor in the fixed dimension functions.

721

Authors: S. Aramuthakannan, R Raja Kumar, G. Mariammal, M. Geetha

In a wireless sensor network (WSN), there are sets of autonomous sensor nodes distributed spatially that use wireless communication to track and document physical or environmental aspects. The series of sensor nodes (SN) cannot be changed when it is installed in an isolated or unattended region due to their wireless nature. Because of the high energy restrictions of wireless sensor devices, it is crucial to carefully manage extreme energy consumption by malevolent nodes in order to enhance network performance. To overcome this challenge fuzzy based adaptive cat swarm optimization for routing (FAR) has been proposed to decrease latency, increase network lives, and minimize energy consumption by reducing the network's energy consumption. There are two stages to the proposed FAR approach. In the first stage, the choice of the cluster head is made depending on things like energy, distance, and transmission cost using fuzzy logic. In the second stage, an adaptive cat swarm optimization method is used to choose the most efficient route for packet routing to maximize node lifetime and to ensure efficient packet routing. The effectiveness of the proposed FAR strategy has been established using metrics like packet delivery, lifetime, and energy efficiency for evaluation. According to the experimental findings, the suggested FAR model consumes less energy than the current FCEEC (fully connected energy-efficient clustering), HSA-CSO (harmony search algorithm and competitive swarm optimisation), and EECHIGWO (energy-efficient cluster head selection using a grey wolf optimization algorithm) models by 43.4%, 32.5%, and 24.1% respectively.

748

Authors: Girish Kumar N G, M N Sree Ranga Raju

In this research, we propose an advanced hybrid precoding design for massive MIMO systems. Our approach integrates two innovative strategies: employing low-resolution ADCs and DACs with non-uniform quantization (NniQ) and implementing dynamic hybrid relay reconfigurable intelligent surfaces (DHRR-RIS). We utilize an iterative alternating minimization algorithm to improve spectral and energy efficiency in the first strategy. The second approach integrates DHRR-RIS with machine learning techniques, including adaptive back propagation neural network (ABPNN) for channel estimation, deep deterministic policy gradient (DDPG) algorithm for hybrid precoding and combining, fire hawk optimization (FHO) for DHRR-RIS, and enhanced fuzzy C-means (EFCM) for data clustering. These methodologies significantly enhance bit error rate (BER) and weighted sum rate (WSR) compared to traditional uniform quantization (UniQ) methods. Our results show that combining low-resolution ADCs/DACs with NniQ and DHRR-RIS, further optimized by machine learning, effectively reduces hardware complexity and power usage while markedly improving BER and WSR, offering a promising direction for efficient massive MIMO system development.

783

Authors: P. Josephin Shermila, V. Seethalakshmi, P. Sujatha Therese, M. Devaki

Foods can be categorized based on their chemical composition, purpose, necessity, concentration, and nutritional value. Protein, fat, and carbohydrates are all categorized as caloric nutrients because they provide the body with the energy it needs to function. In this paper proposed a novel deep learning-based food items classification (SOOTYDEEPFIC) method to classify the variety of food items from the dataset with their calorie values (CV). In the first, the sigmoid stretching methods the images are processed utilized to increase the image quality and remove the noises. Consequently, the segmented images are pre-processed utilizing U-net algorithm. The extracted features are then normalized utilized the sooty tern optimization algorithm (STOA). Food items are classified using Ghostnet based on these relevant features. The experimental result shows the proposed SOOTYDEEPFIC yields an overall accuracy is 99.89%. As compared to existing methods, the proposed SOOTYDEEPFIC shows better results in terms of Accuracy. The Proposed approach enhance the overall accuracy of the proposed SOOTYDEEPFIC, F-RCNN, AFN, and DL applications used to classify fruit images is 99%, 98.8%, 98% and 97 % respectively.

802

Authors: Enas Abdalhussein, Nuha Jameel Ibrahim, Yossra Hussain Ali

Image steganography is a secure communication technique that conceals confidential information within digital images, commonly used in military, digital, and intelligence forensics applications. It preserves data integrity and confidentiality by avoiding attention and suspicion. The goal of this technique is to select random positions within the image to hide data and enhance spatial image steganography. This paper presents a hybrid model based on the salp swarm algorithm (SSA) and particle swarm optimization (PSO). The model is used to locate the optimal position of a text message within a cover image and embed it, ensuring that the stego-image created is of high quality and resistant to specific image processing attacks. The first step involves using SSA to identify the finest pixels from the cover image. In the subsequent step, PSO is employed to choose the best possible bits from these pixels for message concealment. The proposed approach is evaluated on a set of images and compared to existing methods. Experimental findings demonstrate that the suggested system surpasses previous approaches. The suggested method achieved a PSNR of 84.32279 dB and an MSE of 0.00022505 on the Lena image, a PSNR of 84.68192 dB and an MSE of 0.00021362 on the Pepper image, a PSNR of 84.53469 dB and an MSE of 0.00022125 on Baboon, and a PSNR of 84.25440 dB and an MSE of 0.00023651 on the Airplane image. These results indicate higher visual quality and better preservation of data integrity compared to existing methods.

827

Authors: Lara Riyadh Alnajjar, A. E. Abdelkareem

The Internet of Underwater Things (IoUT) systems using Autonomous Underwater Vehicle (AUV) have significant issues with data collecting latency. This work explores the utilization of heuristic approaches and reinforcement learning. Specifically, we employ ant colony optimization (ACO) and Q-learning methods by implementing them using NS3 simulation. Based on the traveling salesman problem (TSP), adopting two objectives, determining the shortest path and achieving a balance between the length of the AUV's tour and increasing the value of information (VoI) of the entire network, moreover, adopting 7 scenarios, every one with a specific style of ordering visiting the sensor nodes. Additionally, the study investigates the integration of ACO and Q-learning algorithms. The results prove the suggested algorithms can obtain the desired paths planning of AUVs dealing with various numbers of SNs (10, 30, and 50) by considering the length of the path, energy consumption, and period of time for data transfer and the computation time. In contrast to previous studies that employed the branch and bound (BB), genetic algorithms (GA), and ant colony algorithm (ACA), the distance obtained is smaller by 7.5%, and the VoI increased by 0.3%. In reference to the examined algorithm's computational time, in ACO using 10 SNs (less by 0.96% for BB method and 0.32% for ACA method, and increased by 0.2% for GA method), however, the utilisation of 30 SNs exhibits a decrease of 0.99%, 52.688%, and 30.37% for BB, ACA, and GA, respectively. Additionally, Q-learning with 10 SNs takes less time than BB, ACA, and GA methods, by 95.4%, 90%, and 82.22%, respectively, as well, with 30 SNs decreased by (99.95%, 97.6%, and 96.47%) for (BB, ACA and GA) methods, respectively.

856

Authors: Lia Farokhah, Riyanarto Sarno, Chastine Fatichah

In the domain of neuroscience, electroencephalography (EEG) holds a pivotal role in determining the inner workings of the human brain, offering real-time insights into cognitive processes, emotions, and neurological disorders. While numerous EEG channels are available in a typical EEG brain-computer interface, selecting the optimal subset for emotion classification poses a significant challenge. Conventional channel selection methods overlook the biological relevance of specific brain lobes in emotional processing, leading to a lack of contextual specificity. This paper introduces a novel approach, by using a biologically informed channel selection approach in the EEG signals. The brain is segmented into various groups and sub-groups and the ability of the channels associated with those groups is determined using time and frequency domain features. The ability of each of these groups and sub-groups to attain higher performance is determined through the accuracy outcomes driven by the support vector machines (SVM). The ability of the selected channels in making accurate classification has been determined using a deep learning model in determining valence and arousal classes, and making a comparison with the selected channels-led classification methods. The approach is validated using the DEAP dataset, demonstrating its potential to enhance EEG-based emotion classification accuracy and efficiency. This innovative methodology offers a promising avenue for future EEG research, allowing customization based on the specific emotions under study, psychological intervention, and streamlining the setup process while maintaining the highest levels of accuracy, reaching an average of 95.7% for intra-subject and 94.65% for cross-subject emotion classification.

880

Authors: Renuka Veerappa Tali, Surekha Borra

The escalating demand for accessible, effective, and precise healthcare solutions has driven extensive research into integrating artificial intelligence into the medical field. Recognizing intricate patterns within biomedical images remains a formidable challenge for human practitioners. In response to this, our study introduces a pioneering image enhancement paradigm, the stationary wavelet channel fusion filtered (SWCFF) algorithm. Additionally, we employ the invariant scattering network (ISN), a novel feature learning methodology to analyze leukocyte images. The novelty of our approach lies in the inventive combination of SWCFF and ISN for enhanced feature extraction and discrimination of leukocytes. Our investigation aims to assess the efficacy of automatically extracted features from this algorithm in differentiating leukocytes, leveraging a support vector machine (SVM) classifier for the diagnosis and detection of leukemia and other blood-related conditions. The proposed model is rigorously evaluated on four benchmark datasets: ALLIDB, C_NMC, BCCD, and LISC. Notably, the ALLIDB binary class achieves a peak accuracy of 96.15% (95% CI: 0.9 to 1), while the class accuracy for neutrophils reaches an impressive 97.96% (95% CI: 0.89 to 1), accompanied by a precision of 100% and a false positive rate of 0%. Our innovative approach holds promise for the deployment of a cost-effective computer-aided diagnosis (CAD) tool in rural settings, aiding physicians in early disease prediction and the timely monitoring of treatment.

913

Authors: Muhammad Aidiel Rachman Putra, Tohari Ahmad, Dandy Pramana Hostiadi, Royyana Muslim Ijtihadie, Pascal Maniriho

Botnet attacks on computer networks require proper handling because they can have dangerous consequences. Botnets are dynamic and able to evolve quickly. A botnet can resemble normal activity, making it challenging to detect. Previous research has introduced botnet detection models but has not focused on analyzing intensity behavior based on incoming and outgoing flows in graph visualization. This analysis is needed to get the botnet attack flow. This paper proposes a detection and comprehensive analysis of botnet attack behavior based on a directed graph. The goal is to detect the attacker and extract the behavior from the directed graph. First, all network traffic is grouped based on the time distance between activities. Visualization is carried out by representing the attacker and target as nodes in every activity group and analyzing the direction of communication in the form of in-degree and out-degree. Meanwhile, interactions are represented in edges and weighted edges based on activity intensity. Then, all graph representation is extracted for classification using random forest, decision tree, support vector classification, Naïve bayes, k -nearest neighbors, logistic regression, and XGBoost. In the experiment, three different datasets are used, namely CTU-13, NCC-1, and NCC-2. The proposed approaches perform well, with an average of 99.97% accuracy, 46.82% precision, and 83.33% recall. These results can form a knowledge base of botnet attacks that can be used in attack detection models on the network.