International Journal of Intelligent Engineering and Systems

1

Authors: Bhupender Singh, Y. D. S. Arya, K. C. Tripathi

Increasing global warming is creating lot of dynamics in polar sea ice extent and concentration. Many works have explored the impact of Arctic sea ice extent over temperature, rainfall in global context. In India specific context, analyzing the impact of Arctic sea ice over Indian Ocean sea surface temperature is critical as the study is important for understanding the monsoon behaviour, aquatic life conditions and effect of storms over Indian coast. This work proposes a novel optimized multivariate long short term memory model for predicting the Indian Ocean sea surface temperature at a fine grained level using the Arctic sea ice extent data. The hyperparameters for optimizing the deep learning model were identified and optimized using Grass hopper swarm optimization algorithm. The model training was itself optimized using hinge loss error minimization. The prediction error is measured in term of root mean square and it was atleast 32% lower in proposed solution compared to artificial neural network and 18% lower compared to convolutional neural network techniques.

24

Authors: Deepika Selvam, Rajeswari Murugesan

In worldwide, COVID-19 has had a significant influence on patients and healthcare systems. Earlier stage of COVID-19 diagnosis and identification are the primary problems in the current pandemic condition. The identification of COVID-19 in CT and chest-X-ray (CXR) imaging is essential for diagnosis, treatment, and evaluation. However, radiologists face a foreseeable issue when it comes to coping with analytical ambiguity in medical imaging. In that situation, a paradigm based on convolutional neural network (CNN) with transfer learning (TL) and taking uncertainty into account was suggested to identify COVID-19 from CT and CXR scan images. However, this method was less capable to extract more useful and distinct image attributes. By fine-tuning the TL network design, this issue can be resolved. The fine tuning model can only fine tune specific layers associated with various goal objectives. However, one of the primary issues with such a method is the selection of layers. To solve this issue, this research uses an enhanced spider monkey optimization (ESMO) technique to select layers of ResNet architecture. Every population of an initialized spider monkey (SM) selects layer and parameter for fine tuning architecture. The fitness value of each SM is used to find best optimal solutions. Categorical cross-entropy loss (CCEL) is considered as fitness of SM. The fitness value of each SM is employed to determine the highest optimal solutions. Subsequent processes such as the stance update process, the learning and decision phase for the local and global leaders of ESMO algorithms, iteratively search for near optimal solutions until convergence. The proposed method can automatically estimate the various CNN layers, which can then be fine-tuned to extract more significant and discriminative features for efficient COVID-19 identification. Finally, the results reveal that the proposed ESMO-ResNet model on SARS-CoV-2 CT database　achieves 91.23% accuracy, which is 21.46%, 20.2%, 12.4%, and 6% higher than the AlexNet, multi-source deep　transfer learning (MSDTL), stacked convolutional neural network (S-CNN) and dynamic mutual training (DMT) models, respectively. Similarly, the ESMO-ResNet model on Covi-19 Radiography dataset achieves 90.06% accuracy, which is 21.34%, 20%, 9.7%, and 3.5% higher than AlexNet, MSDTL, S-CNN, and DMT models, respectively.

46

Authors: Manaswini Pradhan, Ioana L. Coman, Subhankar Mishra, Thanh Thieu, Alauddin Bhuiyan

Early detection of lung cancer in patients can decrease the mortality rate and helps in early diagnosis. An efficient lung cancer prediction system is proposed in this paper known as a long short-term memory (LSTM) based modified remora optimization algorithm (MROA) with multi-objective criteria. Although the MRO is proposed by several researchers, it was based on single-objective with traditional objective functions which tried to maximize the accuracy. This led to the biased classifier with high accuracy and sacrificed sensitivity which resulted in insufficiency of one class and prevalence of other class. To overcome this limitation, a Multi-objective MROA (MMROA) for hyperparameter tuning is proposed, where both accuracy and sensitivity are equally considered as a fitness function. Initially the histopathology images of Lung and Colon 25000 (LC25000) dataset is normalised using colour normalization and segmented with saliency driven edge dependent top-down level set (SDREL) method. The features are extracted using Grey Level Cooccurrence-Matrix (GLCM) and GoogleNet followed by the feature selection using enhanced grasshopper optimization algorithm (EGOA). The selected features are optimized (hyperparameter tuning) using MMROA and fed to LSTM classifier. The proposed model is compared with the existing models such as convolutional neural network (CNN) and enhanced grasshopper optimization algorithm based random forest (EGOA-RF) and obtained remarkable results with accuracy, precision, recall, and f-measure values of 99.02%, 99.17%, 99.03%, and 99.24% respectively.

72

Authors: Amira Hamdi Shaaban, Saleh Mesbah Elkaffa, Gamal Abd El-Nasser A. Said, Osama M Badawy

The identification of selfish mining (SM) activity is critical in blockchain mining. In exchange for successfully mining blocks, volunteers known as miners maintain the network (N/W). Assigning transactions to a block requires verifying each one, then adding the block to the blockchain and broadcasting it to the peers. This paper investigates the profitability of Bitcoin block-discarding attacks. To represent the blockchain’s dynamics, a discrete event simulator is built. A simple N/W concept is employed by everyone as a miner. It is assumed that all miners are honest and that all block broadcasts are latencies, to begin with. In the case of blockchain, the latency produces splits, which are rapidly resolved. The proposed N/W model’s simulation findings closely match the real-world Bitcoin N/W observations. According to the selfish-mine block discarding attack, a small group of cooperating miners can corrupt the Bitcoin system. Both immediate block transmission and latencies block are used to test this claim. This research suggested to use of metaheuristic algorithms such as ant colony optimization (ACO) and particle swarm optimization (PSO) to the mining process and applies them to previous solutions of SM detection to runtime efficiency and solution quality performance and contribute to reducing SM behavior all that by determine the optimal threshold α, optimal hash powering, and the optimal execution time to gain profits with the maximum relative revenue. Finally, our findings indicate that, when optimized, SM performs better than ordinary SM in terms of relative revenue and confirmed blocks. SM is profitable when the threshold is 0.6 and the results appear that the ACOSM optimization is better than PSOSM optimization.

101

Authors: Rawia Abdulla Mohammed, Maisa'a Abid Ali Khodher, Ashwak Alabaichi

The rapid development of IOT and the ease of capturing and transmitting digital images have increased the demand for image encryption. Despite the availability of various encryption methods, chaos-based image encryption is the most suitable method for image applications due to its sensitivity to initial conditions and control parameters. Many image encryption schemes have been developed to enhance the security of images by using chaotic maps, but they cannot combat all possible threats from the cryptanalysis community. This study presents a new chaos-based substitution box (S-box) and its application for securing images. The construction of the presented S-box approach is based on a new five-dimensional hyper-chaotic system, and the experimental results prove that the suggested S-box approach has high nonlinearity and good cryptographic characteristics. In addition, we utilize the effectiveness of the presented S-box approach in designing a novel lightweight image encryption/decryption scheme. Results demonstrate that this scheme has a sufficient peak signal-to-noise ratio, a low correlation, and a large key space. These factors make it more efficient than its classical counterpart and allow it to resist statistical and differential attacks. The number of changing pixel rate (NPCR) and the unified averaged changed intensity (UACI) values were 99.623 and 33.532, respectively. The entropy oscillates from 7.9965 to 7.9971 for the tested encrypted images. These criteria indicate an efficient, fast, and robust scheme that can be used for secure image transmission via public channels.

123

Authors: MR Archana, Dayanand Jamkhandikar, Deepak N. Biradar

Protecting the data against forgery is an important concept and digital images are necessary for exhibiting information. Digital image forgeries are attaching extraordinary patterns to original images and it causes visual heterogeneousness. Image copy-move forgery is a challenging technique, that involves copying part of an image and then pasting the copied part into the same image. In this paper, the golden jackal optimization (GJO) is proposed for feature selection and multi-support vector machine (M-SVM) is proposed for effective classification. The GJO optimizes the feature selection process by iteratively searching for the most informative subset of features. By leveraging the exploration and exploitation of GJO, the algorithm efficiently explores the feature space, selecting relevant features that contribute to accurate forgery classification. This enhance the performance of MSVM by focusing on the most discriminative features. For this work, a dataset named as MICC-F2000 is examined which is managed by 2000 images in this 1300 are original and 700 are forged images. The result shows that the proposed GJO based MSVM model delivers the performance metrics like accuracy, sensitivity, specificity, precision, and MCC values about 99.47 %, 97.01%, 96.51%, 99.62%, and 96.39% respectively which ensures accurate forgery detection compared with existing methods such as SSDAE-GOA-SHO, ConvLSTM, CNN and dual branch CNN methods.

145

Authors: Keerti Naregal, Vijay Kalmani

The internet of things (IoT) faces significant obstacles due to insufficient identity recognition and evolving network architecture, leading to concerns about the confidentiality of data and causing anxiety. The attribute-based encryption (ABE) techniques have recently been considered a solution to guarantee the security of data transfer and precise data sharing. However, most of the existing methods used the attribute-based encryption (ABE) technique, which requires a lot of computation power and is unsuitable for IoT devices with minimal resources. Researchers have achieved improvements in establishing practical methods for cloud security on mobile IoT devices using lightweight ABE. In this paper, the ciphertext policy-revocable and searchable attribute-based encryption (CP-RSABE) method is proposed to protect privacy and security. The proposed methods greatly lower the cost of computing IoT devices with the availability of multiple-keyword searchers for the users of data. The user’s side of computation is very efficient, and the cloud server handles most of the computing tasks. The proposed method performs significantly better in terms of ciphertext size, decryption time, and parameter size. The method achieves data security, privacy preservation, and mobile terminal operations that are suitable for applications of IoT methods. The existing methods such as online/offline multi authority-ABE with cryptographic reverse firewalls (OO-MA-ABE-CRF), ciphertext policy ABE (CP-ABE), ABE with full privacy protection (ABE-FPP) are used to justify the effectiveness of CP-RSABE method. The proposed method CP-RSABE achieves the encryption time (0.0163s), decryption time (0.25s), communication overhead (3.4KB), size of secret key (5.1KB), and size of ciphertext (10.7KB) compared to the OO-MA-ABE-CRF, CP-ABE, ABE-FPP.

171

Authors: Liqaa Saadi Mezher, Ayam Mohsen Abbass, Basma Jumaa Saleh

This paper introduces a novel approach to enhance data security through the combination of the caesar cipher with triple pass encryption, utilizing the Krill herd optimization algorithm. The proposed method aims to safeguard sensitive information during transmission and storage by leveraging the strengths of both encryption techniques and optimizing the generation of random keys. In this approach, the caesar cipher adds an extra layer of security by shifting each character in the plaintext by a fixed number of positions. The triple pass encryption further reinforces the encryption process by applying the caesar cipher three times, each with a different shift value. To evaluate the effectiveness of this combined approach, it is compared against the basic caesar cipher and basic triple pass protocol. The experimental results demonstrate superior performance of the proposed method in terms of minimal processing time, 100% accuracy, and zero errors when compared to these traditional techniques. The Krill herd algorithm plays a crucial role in generating random keys and determining the optimal shift values for both the caesar cipher and triple pass encryption. The effectiveness of this combined approach is compared to existing techniques, and the experimental results demonstrate superior performance in terms of minimal processing time, 100% accuracy, and zero errors. This promising solution holds potential for enhancing data security across various domains, as the Krill Herd-based key generation significantly improves the randomness of encryption keys, thereby increasing their resilience against cryptographic attacks.

198

Authors: Ziyad Tariq Mustafa Al-Ta'i, Shaima Miteb Sadoon

Data security and accessibility in a cloud computing environment are successfully guaranteed by the safe privacy content-based image retrieval (CBIR) approach. However, ineffective cipher text methods and feature-extraction techniques might reduce performance. Therefore, in order to create secure privacy CBIR, it is still necessary to address the need for high-security encryption techniques and the capacity to extract characteristics from cipher text images. To achieve this, we provide a secure privacy image retrieval method based on convolutional neural network features. Initially, a 3D Lotka_Volterra chaotic systems encryption method based on visual cryptography encodes images, and the updated DenseNet-121 model is fine-tuned by employing the encrypted images to construct a feature extractor. There are multi-authorized users set up for the fine-tuning feature extractor. The encrypted features and images are uploaded to a cloud server. Experiments on the Corel10k dataset showed the effectiveness of the proposed model in terms of security and precision compared to previous works, achieving an average search precision of 68.94% and an MSE =12764.27263, PSNR = 7.300480125 dB between the original image and encrypted, and the decrypted image is entirely identical to the original image.

225

Authors: Udayakumar K, Ramamoorthy S

An intensive computation source has become increasingly important in recent years to meet the intensive resource and low-latency needs of industrial internet of things (IIoT) systems. Existing IIoT devices are built with limited computational resource, delivering results in a limited fashion when used in highly resource-intensive and delay-sensitive applications. It is difficult to process time-critical IIoT task due to varying demand like low latency, intensive computation and high data transmission. Offloading computing tasks to mobile edge computing (MEC) servers in the network's perimeter can effectively reduce delay. However, MEC server collected fewer resources than the resource cloud. To improve the resource utilization and minimize cost, this research develops an adaptive task offloading decision model through multi-constraint objective function. The goal is to minimize service delay, energy consumption, and maximize resource utilization through prediction based decision model. This study examines a non-orthogonal multiple access (NOMA) -based MEC for IIoT system, where edge nodes offload their tasks to nearby edge servers for execution. Heuristically modified long short-term memory (H-LSTM) employing hybrid cat and mouse dingo optimization (HCMDO)-based reinforcement learning is suggested to distribute tasks optimally. We formulate joint optimization by considering multiple parameters using HCMDO. Further, these optimal parameters are used in training H-LSTM along with benchmark dataset. The outcome of the H-LSTM network utilized in deep reinforcement learning (DRL) to improve convergence speed, accuracy and stability by predicting task and best server. Average energy consumption analysis performed in the developed model attained 19.8%, 15.1%, 16.9%, and 15.6% than conventional approaches. In addition, the experimental results shows developed model attain better outcome in terms of delay and resource utilization.

249

Authors: Samer Saeed Issa, Abdullah Hasan Ali, Sinan Q. Salih, Faris Hasan Taha

This paper introduces a novel approach for generating strong substitution boxes (S-Boxes) using the black hole algorithm (BHA) integrated with Arnold chaotic map (ACM) and Henon chaotic map (HCM). The ACM is employed for enhanced initialization of the BHA, while the HCM is utilized for generating new stars during the searching process. The proposed algorithm, named BH-SBOX, aims to enhance the nonlinearity and cryptographic properties of the generated S-Boxes. Experimental evaluations further demonstrate that the BH-SBOX algorithm produces S-Boxes with excellent nonlinearity, strict avalanche criteria, bit independence criteria, differential uniformity, and the maximum expected linear probability. The proposed method demonstrated a high nonlinearity property, achieving a value of 108.25, which is considered successful and effective. This exceptional nonlinearity can be attributed to the utilization of the Henon map, which enables good exploration of the solution space during the search process. The chaotic and unpredictable nature of the Henon map contributes significantly to the algorithm's ability to discover S-Boxes with enhanced nonlinearity and robust cryptographic characteristics. The average bit independence criteria (BIC) and strict avalanche criteria (SAC) were found to be 102.85 and 0.50392, respectively. These results indicate that the proposed method successfully generated S-boxes with strong cryptographic properties, ensuring robustness and resistance against various attacks.

274

Authors: Syeda Ambareen Rana, Bharati Harsoor

The creation of an ad hoc network suffers from technical barriers due to its limited energy source, unpredictable wireless connectivity, and dynamic network topology, which is a challenging concern in predicting link quality among the nodes. Therefore, it is essential to predict link trade-offs between the nodes in such a dynamic network and also preserve the energy resource to have a better link and longer life of the network. This paper presents an efficient link cost prediction method (LCPM) for dynamic route switching based on the node energy source. The LCP method determines the stable and minimum link cost route to minimize the packet loss which might cause due to node mobility and link failure in reactive routing protocols. It initially suggests a mechanism to predict the least energy utilization route between source and destination to reduce the routing overhead. Later, it defines the methodology of establishing the least link cost route using LCP to achieve better throughput and low packet loss. The experiment evaluation results through varying nodes scalability show the effectiveness of the proposal. It shows the improvisation of LCPM through achieving an average of 3.35% better PDR, 2.93% of lower routing overhead and 8.105J lower energy consumption. It also achieves an average of 90ms with 20 nodes and 50ms with 100 nodes lower E-2-E delay in compare to the existing methods.

298

Authors: Jyothirmai Joshi, Malini Mudigonda, Raghava Morusupalli

Retinal disease is a potentially fatal condition that causes blindness, therefore early prediction of retinal disease is an effective way to minimize the problem of blindness. Optical coherence tomography (OCT) is an efficient technique for scanning precise images of the retina to diagnose the macular holes and to view the age-related macular extension. A tournament based levy multiverse optimization algorithm (T-LMVO) is proposed in this research to identify macular pathologies with efficient feature selection of retinal images. Although the TLMVO was proposed in existing researches, the robustness of the local optimization and the capacity of global optimization is minimum. To improve the robustness of these optimizations, LMVO is enhanced with the tournament based feature selection which has an advantage of handling minimization problems and can continue to update strongly even at the end of iterative phase. Overall, the performance of the proposed accuracy based T-LMVO shows better stability, and robustness. The process of feature selection using T-LMVO resulted in improved performance in the retinal OCT images, which have the accuracy of 98.95%, sensitivity of 99.02%, specificity of 97.61%, MCC of 98.92% and an error rate of 1.05 and noor eye hospital images dataset has accuracy of 98.99%, specificity with 99.38%, matthew’s correlation coefficient (MCC) with 96.95%, sensitivity with 97.63%, and the error rate was reduced to 1.01%.

328

Authors: Rakshith Nagaraj, Minavathi

A wireless sensor network (WSN) is a collection of nodes fitted with small sensors and transceiver elements that are utilized for sensing, tracking, and data collection in a variety of situations. Sensor energy is considered a significant resource due to the difficulty in providing battery a backup to sensors when they are deployed in an uncontrolled or wild area. Integrating clustering & routing strategies are the optimal techniques for saving sensor node energy. Cluster heads (CHs) are selected carefully for load balancing throughout the clustering process. Therefore, multi-objective sand cat swarm optimization (MSCSO) algorithm is proposed to maximize the efficiency of energy in WSN. The optimum cluster heads from the network are selected and the path through the cluster head is identified by utilizing MSCSO. The proposed MSCSO decreases the usage of energy of nodes when maximizing the transmission of data in the WSN. The performances of MSCSO are evaluated based on alive nodes, energy consumption, throughput, and life expectancy for 100 nodes and attained better performance than other methods like low energy adaptive clustering hierarchy (LEACH), distributed energy efficient clustering (DEEC) and threshold-distributed energy efficient clustering (T-DEEC). The MSCSO is compared with existing methods such as fractional artificial lion (FAL) algorithm, taylor-spotted hyena optimization (Taylor-SHO) algorithm, energy centric multi objective salp swarm algorithm (ECMOSSA) and whale based tunicate swarm algorithm – aquila optimizer (WTSA-AO) and shows that the MSCSO algorithm improves the performance by reducing energy consumption of 0.0208J, 0.0350J and 0.04045J for 200, 400 and 600 rounds respectively.

356

Authors: Kowsalya Selvaraj, Saraswathi Subbian

Accurate early detection and diagnosis of hepatocellular carcinoma (HCC) need a patient similarity network (PSN) to increase the patient’s survival rate. For this reason, the PSN with a dense graph convolutional neural network (pDenseGCN) model has been developed for liver cancer diagnosis. But an over-smoothing problem occurred in the GCN when increasing the network depth, resulting in poor classification performance. Therefore, this article proposes a novel inter-scale amalgam cluster DenseGCN (IACDGCN) model to tackle the over-smoothing problem and increase the accuracy of HCC detection. The major processes in this model are the following: (i) first, the similarity network fusion (SNF) and denoising auto-encoder (DAE) are used to create PSN and extract a latent embedding representation of multi-omics data, respectively; (ii) then, the obtained PSN and extracted feature matrix are given to the IACDGCN model for classification. This IACDGCN is built by comprising a cluster DenseGCN unit and a spatiotemporal attention unit with a larger receptive field (RF) to dynamically strengthen the intermediary feature maps. Also, an inter-scale combined temporal convolution unit is used to learn discriminable temporal feature maps via the mixture of various scale convolution kernels. Moreover, those feature maps are passed to the softmax classifier to get the probability of each class such as healthy and HCC patients. Finally, the experimental results show that the IACDGCN model realizes 96.6% accuracy which is higher than the random forest, artificial neural network (ANN), deep neural network (DNN) and DenseNet models on the liver hepatocellular carcinoma (LIHC) omics dataset.

375

Authors: Ntivuguruzwa Jean De La Croix, Tohari Ahmad, Royyana Muslim Ijtihadie

Concealing confidential data within digital multimedia, also called steganography, has been crucial for safeguarding data. However, steganalysis techniques are best known for compromising these data concealment methods. The convolutional neural network (CNN) yielded better performance over the previously suggested techniques based on machine learning for steganalysis. Although the current approaches show promising outcomes, they encounter challenges in locating the secret data. This paper introduces a CNN architecture with a key idea to enhance the accuracy of identifying the precise hidden data location. Our method novelty is based on reducing image features and network parameters via customized feature allocation and convolutional kernels optimization and enhancing classification accuracy by optimizing the kernels combined with back-forward gradient descent. Our experimental results show improvements over the existing works with the accuracy under payload capacity of 0.4 bpp from 83.01% to 98.78% for S-UNIWARD and 87.19% to 97.93% for WOW.

398

Authors: B Ramakantha Reddy, R. Lokesh Kumar

With the increasing complexity of contemporary E-commerce recommender systems, personalized health product recommendations have become a challenging task. Existing methods predominantly rely on latent natural language features extracted from product descriptions, lacking transparency and user engagement. While recent item-representation-learning algorithms integrate pre-defined attribute data, they still struggle to provide accurate recommendations and detailed explanations due to sporadic relationships between stated qualities and the recommendation process. To address these challenges, we propose an E-commerce based personalized health product recommendation system using a CNN-Bi-LSTM model. Our system leverages pre-trained CNN-based transfer learning models, such as AlexNet, Google Net, and ResNet-50, for predicting health products, while the Bi-LSTM acts as a numerical function to provide ratings. The CNN-Bi-LSTM recommendation system utilizes attribute-specific representation extraction methods, resulting in a more robust compatibility model and improved recommendation effectiveness for health products. The proposed approach is evaluated on two datasets: The health-product dataset and the Flipkart health product dataset. The health-product dataset comprises 20,726 products, including health supplements (14,871 items) and fitness-related products (13,663 items) recommended by experts. The Flipkart health product dataset contains 21,889 healthcare products with essential information like name, price, MRP, discount, and ratings. Comparative analysis against existing methods, such as W-RNN and Improved SVM, demonstrates the superiority of our CNN-Bi-LSTM model. The evaluation metrics, including accuracy (95.52%), recall (88.31%), and user coverage (89.51%), highlight the effectiveness of our recommendation system in providing more accurate and relevant personalized health product recommendations. For the aforementioned dataset, the AUC and hit rate of the proposed method are contrasted with those of existing methods. The findings were gathered to demonstrate that the suggested approach performs better when compared to the current recommendation systems.

422

Authors: Suhaila N. Mohammed, Adnan J. Jabir

Infants typically cry to get their parents’ attention. Through their cries, infants express their basic needs like hunger, tiredness, pain, and discomfort. Unfortunately, it is difficult to interpret cries to comprehend the demands of an infant. The only way to solve this problem is to analyze the infant's acoustic speech pattern and determine the cause of the crying. In this study, the cry signal is converted to a spectrogram image to take advantage of the wide spectral range of image-based features. Before generating the represented features, the watershed segmentation algorithm is used to remove distracting areas of the image. Then, histogram of gradients (HoG) features are generated. Because the feature vector has high dimensionality, two stages of dimensionality reduction are presented. First, the feature pool is decreased using the fisher score feature selection approach. The ideal feature set is then chosen using a combination of transfer learning, genetic algorithm (GA), and neural networks. To motivate GA to pick characteristics that will operate successfully with the neural network, a ranked aware mutation operator is suggested. As system evaluation material, the donateacry-corpus public dataset is employed. Experiments reveal that when 80 HoG features are generated and the best 37 Fisher scores are chosen, the model has the best accuracy of 92% when applying transfer learning to 11 hidden layers of the neural network. The study's findings support the use of image-based features to identify the cause of a baby's crying.

445

Authors: Akash Rathor, Manoj Kumar, R. K. Mishra, Shivender Goswami, Ankit Chaudhary

Artificial intelligence and machine learning, as intelligent systems, have a vital role to play in both the development and deployment of solutions for post-quantum cryptography. Post-quantum cryptography's (PQC's) most important aspect is cryptography based on isogeny (IBC). It is extensively used because of its compatibility and shorter key lengths. Point operations and isogeny computations are used as the basic building blocks in the implementation of the IBC. SIDH (supersingular isogeny diffie-hellman) and CSIDH (commutative supersingular isogeny diffie-hellman) finds application in isogeny-based cryptography for the provision of secure key exchange protocols and cryptographic primitives. The isogeny degree holds great significance in both SIDH and CSIDH schemes, serving as a critical parameter that profoundly impacts the security and efficiency of these cryptographic protocols built upon isogenies. In actual IBC implementations, where decreasing computational overhead can greatly enhance system performance. Montgomery curves are used in the literature because they can be used to carry out a specialised point operation. Methods for calculating 2, 3, and 4 isogenies on the Huff curve have been proposed in the current study. These techniques include changing an affine plane into a projective plane. In terms of computational cost, the study discovered that the suggested methods for computing isogenies on the Huff curve are more effective than those utilising the Edwards and Montgomery curves.

470

Authors: Dhivya Karunya Sampath, Krishna Kumar

In crowded environments, the importance of automatic video surveillance cannot be overstated. It plays a vital role in detecting unusual incidents and averting accidents, particularly in areas teeming with pedestrians. Surveillance frameworks demonstrate their inherent strength when deployed in real-world scenarios. In this paper, an efficient algorithm that can identify abnormalities in videos has been proposed as a solution to the problems of high computing power and variation between negative and positive samples. To address the issue of fewer negative samples, the algorithm employs the spatiotemporal inter-fused autoencoder in a method of unsupervised learning to locate and extract the samples of negative from the dataset. A spatial-temporal convolutional neural network (CNN) is built using this methodology, and it has a basic structure and requires minimal computing power. To develop the model of detection, the spatiotemporal CNN is trained using the method of supervised training with negative and positive data. The UMN and UCSD datasets are used as benchmark datasets for this method. The outcomes of the experiment demonstrate that the proposed method is more accurate than the existing algorithms at both frame and pixel levels and it can locate anomalous behaviors in real-time. The proposed method achieves better accuracy of 99.76%, 99.92%, 99.15%, 98.45%, and 94.67% in UCSD ped1, UCSD ped2, and UMN scene 1, 2, 3 datasets compared to the hybrid CNN and RF classifiers, MCMS-BCN attention+densenet121/Efficientnetv2, gradient motion descriptor (PGD) and enhanced entropy classifier, and attention mechanism. Additionally, the proposed method achieves better AUC of 99.96%, 99.83%, 99.97%, 90.15%, and 99.72% in the datasets of UCSD ped1, UCSD ped2, and UMN scenes 1, 2, 3 compared to the low-rank and compact coefficient dictionary learning (LRCCDL), and hybrid CNN and RF classifiers.

492

Authors: Jayaram C V, B K Raghavendra

In the videos, co-location (CL) pattern recognition (PR) aids the systems administrators in effectively recognizing abandoned and detached co-located objects in the airport terminals. Nevertheless, for co-located object recognition (OR) in video streams, works are developed scarcely. Hence, this paper proposes a co-located OR with movement analysis based on mullers’ decomposition-based principle component analysis (MD-PCA) and varied distance index-based farthest first clustering (VDI-FFC). Frame conversion, pre-processing, and object detection with you only look once version 8 (YOLO v8) are performed to detect the objects in the input video. Next, to determine the main and co-located objects in the frame, the features from the detected objects and reference images are extracted and matched. Afterward, to determine the pattern movement pattern, the motion estimation with rood pattern search (RPS) and pattern estimation with random walk-normalized laplacian-based hough transform (RWN-HT) is performed with the filtered objects frame. Next, the distance is estimated and with the estimated distance, the TarIf activation-based long short term memory (TI-LSTM) predicts whether the co-located object is detached or abandoned by the main object. If the object is detached/ abandoned, such objects are recognized using the MD-PCA, also, to track the location of co-located objects, movement analysis is performed with VDI-FFC. Further, the alert is sent to the administrator monitoring the video surveillance in the airport terminal based on the determined position. If a co-located object is discovered to be disconnected, the surveillance administrator is informed. The user may be informed by the administrator through a public announcement. In 6017 milliseconds, the proposed MC-PCA achieved a recognition rate of 96.75%. With VDI-FFC, the movement analysis of the associated item is completed in 334 milliseconds. In 4535ms, the RWN-HT evaluated the patterns. With 97.26% accuracy, 98.81% precision, and 0.0675 MAE, the TI-LSTM identified the availability of objects.

519

Authors: Sura Mazin Ali, Alyaa hasan zwiad, Rusul Mansoor Al-Amri, Alaa Kadhim Farhan

The continuous development in the digital world and the significant increase in the use of digital images have made it important to guarantee protection in the digital world. In the field of security, the advancement of cryptographic algorithms is essential. Recently, cryptography relying on chaotic systems has been suggested as a reliable and effective approach for image protection. Chaos systems have been widely used in security applications due to their good features such as being unpredictable, dynamic, and sensitive to starting conditions and control parameters. The paper presents a new idea for a color image enciphering scheme and creates S-Box based on the 2D-HSM chaotic map. The main contribution of this scheme is how to create and design a new S-Box against many attacks methods. The designed S-Box succeeds all the S-box test standards such as balanced, completeness, avalanch, and strict avalanch. Besides that, enciphering work evaluation parameters such as information entropy, correlations coefficient, histograms, NPCR, and UACI are collected. According to the findings, the entropys, NPCR, and UACI scores obtained for the baboon image are 7.996, 99.64%, and 33.57%.

542

Authors: Haider D. Albonda

5G mobile networks provide two main uses: enhanced mobile broadband (eMBB) and ultra-reliable low-latency communications (uRLLC) are two critical services. While eMBB focuses on maximum data rate, uRLLC prioritizes latency and reliability. However, sharing radio resources between these two use cases creates a complex planning problem due to the trade-off between these different factors. Our proposal introduces an efficient control strategy that optimizes the use of licensed and unlicensed radio resources across various services i.e., eMBB with guaranteed bit rate (eMBB-GBR) and uRLLC ensuring that the traffic requirements of each service are met. The main objective of the proposed control strategy is to maximize the utilization of resources by ensuring the availability of resources, reducing delays, and achieving high throughput. The simulation results clearly show the superiority of the proposed algorithm compared to the reference scheme. This significantly improves the performance of the network by improving the utilization of resources, reducing latency service, increasing the rate of successful packets, and reducing the probability of blocking. The algorithm shows significant improvements, particularly with an 8%, 22%, and 28 % increase in effective capacity for the eMBB-GBR service compared to the reference approaches when the session arrival rate is 1.2. In addition, it significantly reduces the probability of session drops for eMBB-GBR service by about 0.05, 0.12, and 0.17 compared to the reference approaches. Moreover, the proposed control strategy has a significant relative latency for uRLLC service of 13%, 21 %, and 26% compared to reference approaches when URLLC density is around 1000 UEs.

564

Authors: Komal R. Hole, Divya Anand

The classification of EEG (Electroencephalogram) signals requires design of multidomain modules, including signal pre-processing, filtering, segmentation, extraction of features from segmented signals, reduction of features via statistical modelling, categorization of the signal into 1-of-N brain disease classes, and performing post processing operations. Researchers have proposed deep learning models with single domain features for representing EEG signals which limits performance capabilities, when used for multiple disease types. In deep learning models feature extraction & selection are wrapped in black-box containers, which is uncontrollable without compromising classification performance. To overcome this, design of multispectral data representation engine for classification via ensemble models used. The proposed engine represents input EEG signals into Mel frequency cepstral coefficient (MFCC), and iVector components. The MFCC feature vector is built using cepstrum, spectrum, power density, and other frequency domain features, while iVector is built using statistical entropy features. This combination of feature sets can improve feature representation efficiency assists in optimizing classification performance. A novel ensemble classification model is designed using multiple neural networks (MNNs) varies layer size with observation that proposed model showcased over 98.5% accuracy for classification. Proposed AMVAFEx model has outperformed than existing models like online transfer TSK fuzzy classifier (TTFC),Neuroglial Network model (NNM) and Local Binary Pattern Transition Histogram (LBP TH) in terms of accuracy, precision, recall & delay performance under different input conditions. With this advantage, proposed model is useful for real-time clinical applications.

590

Authors: Lavanya Nagichetty Lakshminarayana, Shashi Rekha Gangadharaiah, Madhushree

The approach to ensuring secure transmission in a mobile Ad hoc network (MANET) involves safeguarding the information exchanged between devices. It involves using encryption and authentication techniques to make sure that only authorized devices can access the data and that the data remains unchanged and trustworthy. However, implementing secure broadcasting can be challenging due to factors like limited resources, the dynamic nature of networks, and trust management. To overcome those issues, the trust-based multi-objective-pelican optimization algorithm (TMO-POA) method is used in this paper to offer great security for the data packets and stop malevolent attacks. The energy efficacy of the node and trust values are evaluated. The superior cluster head is selected for that TMO-POA appraisal of the cost function. Let's detect the malevolent node in the network, decrease the packet loss, and evaluate the trust value of the node in its neighbourhood. Upgrading the routing detail is made easier by the trust value and maximizes the network throughput. The outcomes attained from the simulation show that the suggested TMO-POA succeeded in excellent execution according to various parameters metrics. From the results, the proposed TMO-POA, succeeded in achieving maximum throughput (550.01 Kbps), Average energy consumption (0.90J), routing overhead (0.0550), PDR (0.999) and delay (11.23ms), which is much better than the existing bacteria foraging optimization algorithm (BFOA), cat slapped solo algorithm (C-SSA), and trust-based efficiency routing (TER) algorithms.

610

Authors: Kavitha Shanmugam, Sarojini Kaliyamoorthi

Due to the variety of fruit disorders, food production is highly impacted, resulting in economic degradation. So, proper disease recognition and classification system are vital to reduce agricultural losses and encourage economic growth. For this purpose, a multi-scale region-of-interest (ROI)-based deep learning (MRDL) model was developed, which localizes and classifies multiple fruit disorders based on the new attention principle and feature representation. But, the shadow regions in the fruit pictures influence the model efficiency and the picture dataset was highly dynamic while adding new classes for different disorders; because collecting diseased class labels for a single fruit picture was not simple. Hence in this article, a novel cycle-generative adversarial network (CycleGAN) model is integrated with the MRDL to generate shadow-free fruit pictures and increase the accuracy of classifying multiple classes of fruit infections. Initially, an attribute registration and shadow removal adversarial network (ArSAN) is proposed to generate various fruit pictures without shadow areas. In this ArSAN, the attributes of objects and shadows in fruit pictures are initially obtained. Then, those are labeled by the attribute enrolment scheme with the CycleGAN. Based on the labeled attributes, the filtering weights are set to all pictures in the real picture dataset. Also, a similarity finding scheme is used to discover the best-matched pictures for the generator network. As a result, multiple generators are trained to produce the shadow-free fruit picture dataset with diverse cleanness characteristics. Moreover, the newly generated pictures are passed to the MRDL model for localizing the infected fruit areas and classifying the multiple disease classes. Finally, the experimental outcomes reveal that the ArSAN-MRDL model attains a 93.8%, 93.78% and 93.76% accuracy on apple, citrus and tomato disorder picture sets, respectively contrasted to the MRDL, artificial neural network (ANN), 13-layer convolutional neural network (CNN), L2MXception, CycleGAN-DenseNet, and conditional GAN (cGAN)-VGG16 models.

638

Authors: Ramya Asa Latha Busi, James Stephen Meka, P V G D Prasad Reddy

A significant global public health concern is the widespread presence of chronic kidney disease (CKD). High mortality rates are associated with the disease, particularly in developing nations. Since there are no visible early-stage signs, CKD frequently goes undiagnosed. In the meantime, preventing the disease from progressing requires early detection and prompt clinical care. To help clinicians discover CKD early, deep learning (DL) techniques can give an effective and affordable diagnosis. This research proposes a unique hybrid DL approach to classify CKD. In a pre-processing step, eliminate the missing values and reduce noise from data, data transformation, and outlier detection. After that, using the improved capsule network (Improved CapsNet) method to extract the features. Then, select essential features using the improved spotted hyena optimizer (ISHO) algorithm to better classification with less time. Finally, employ hybrid deep learning techniques of BConvLSTM and DNetCNN to classify the CKD. A recently introduced CKD prediction algorithm and well-known classifiers were used as benchmarks for the proposed approaches. The proposed model, which was trained with the smaller feature set, outperformed other classifiers with a classification accuracy rate of 99.89%. The experimental findings also demonstrate the positive effect of feature selection on the performance of the different techniques. The proposed technique has developed a reliable predictive system for recognizing CKD and may be extended to more unbalanced medical datasets to identify diseases reliably.

659

Authors: Zaid S. Mohsen, Mohamed Jasim Mohamed

In this paper, four control structures are provided for continuous ctirred-tank reactor (CSTR) system's fractional/integer order proportional integral derivative neural network controllers. The revised neural network weights and the controller's parameters are optimized using the optimization technique called ant colony optimization (ACO). The proposed controllers' resistance to changes in the initial state, outside disturbances, and parameter modifications is also tested. The fractional order proportional integral derivative neural network controllers provide the best assurance and also enhance the system's robustness to changes in the initial state, external disturbances, and parameter variations, according to the results of MATLAB code.The fractional order proportional integral derivative neural network controller1(FOPIDNNC1) is the best structure among all those with the minimum cost function equal to 0.011588 for the set-point variations, 0.015325 for uncertainty parameter, 0.018274 for disturbances rejection, and the best structure fractional order proportional integral derivative neural network controller3 (FOPIDNNC3) among all those with the minimum cost function equal to 0.008733 for tracking.

684

Authors: Abdelrahman Hussein, Qusai Shambour

In this era of vast online healthcare information, patients often find it challenging to choose the most appropriate doctors from a myriad of options available on different communities. The sheer volume of choices can be overwhelming, underscoring the importance of personalized doctor recommendations. Unfortunately, many existing communities rely on generic doctor recommendations through a global ranking system, potentially neglecting the unique needs and preferences of individual patients. To address these challenges, this study introduces a novel multi-criteria user-item trust-enhanced collaborative filtering (MCUITeCF) approach, specifically designed to facilitate patients in locating doctors who match their unique preferences. This proposed approach capitalizes on multi-criteria ratings and integrates user-item trust relationships, aiming to enhance the quality of recommendations, while tackling the common issues of data sparsity and the cold-start problem. An examination of the proposed approach, using a real-world healthcare multi-criteria dataset, the RateMDs dataset, reveals its effectiveness in overcoming challenges associated with data sparsity and cold-start problems. The results demonstrate improved prediction accuracy and coverage compared to benchmark approaches, namely MC user-based CF, MC item-based CF, MC semantic-based CF, MC trust-based CF, and trust-semantic enhanced MC CF. Specifically, the results indicate that the MCUITeCF approach improves the average MAE by 66% compared to all of the benchmark approaches when tested on the RateMDs dataset. When dealing with data sparsity, the MCUITeCF approach improves the average MAE by 41% and prediction coverage by 23% compared to all of the benchmark approaches. In scenarios involving cold-start items, the MCUITeCF outperformed specific benchmark methods such as MC item-based CF, MC semantic-based CF, and Trust-Semantic enhanced MC CF, registering a 21% drop in average MAE and a 28% rise in prediction coverage. Similarly, for cold-start user situations, MCUITeCF excelled by decreasing the average MAE by 29% and a substantial increase in prediction coverage by 20% compared to MC user-based CF, MC trust-based CF, and trust-semantic enhanced MC CF benchmark approaches.

706

Authors: Amol Subhash Dange, Manjunath Swamy Byranahalli Eraiah, Asha Kethaganahalli Hanumanthaiah, Manju More Eshwar Rao, Sunil Kumar Ganganayaka

Web crawlers gather and analyze a large amount of data available online to obtain specific forms of objective data, such as news. Web crawlers are becoming more important since big data is used in numerous different sectors and web data is rising dramatically each year. However, when analyzing large volumes of information and making rapid decisions, the organization frequently uses minimal data, which leads to inefficient choices. In this paper, the minibatch stochastic gradient descent (SGD) optimization and radial basis function SVM are proposed to assist organizations in the targeted crawling of relevant online artifacts and semantically matching them against internal big data for better strategy decisions. The proposed method has been used and extensively evaluated in the e-procurement field. The minibatch SGD optimization and radial basis function SVM has gradually been expanded to include more fields such as robot programming and cloud hosting. The existing methods of web crawler for pharmacokinetics (WCPK), automatic extraction method, malicious website detection techniques, and BERT with softmax layer method are used to justify the effectiveness of the proposed minibatch SGD optimization and the radial basis function SVM method. The proposed method achieves better precision, recall, and f1-measure of 99.25%, 98.91%, and 99.57% on DMOZ dataset and 96.23%, 94.71%, and 97.53% on synthesized dataset when compared to the existing methods.

731

Authors: Purva Yogesh Kamat Mhamai, Suraiya Tarannum

Nowadays, modern mobile telecommunication systems use multiple input multiple output (MIMO) with orthogonal frequency division multiplexing (OFDM) because of its robustness and huge spectrum efficiency. The most significant issues in MIMO-OFDM are noises occurred during communication and precise channel estimation. In this paper, the double density dual-tree complex wavelet transform (DDDTCWT) based denoising is proposed for eliminating noises occurred over the MIMO-OFDM. The fast independent component analysis using negentropy (FICAN) is used for performing blind channel estimation according to the statistics obtained from the received signal. Therefore, an effective denoising and channel estimation accomplished in the MIMO-OFDM is used to reduce the errors during the transmission. The performance of DDDTCWT-FICAN is analyzed using mean square error (MSE), bit error rate (BER) and symbol error rate (SER). The existing research used to evaluate the efficiency of DDDTCWT-FICAN are pilot-based interpolation (PBI) technique, adaptive optimized fast blind channel estimation (AOFBCE) and discrete fourier transform (DFT) with denoise model based least square (LS) Wiener namely DFT-LS-WIENER. The BER of the DDDTCWT-FICAN for 10 dB of SNR is 0.0053, it is less when compared to the PBI.

753

Authors: Shivanand V. Manjaragi, S. V. Saboji

The fifth generation (5G) networks are a popular standard that carries effective skills to conquer the tests of next generation wireless networks. Also, the 5G systems can support high data traffic by rendering high throughput and low latency towards the massively connected nodes. Here, handover is highly significant for data processing, portability and real time data creation in mobile technologies. With the 5G entrance, the cellular network has become a completely heterogeneous network (HetNet). The Software Defined Network (SDN) concept is used in 5G HetNets for better mobility management. Most existing research works have concentrated on handover authentication, but those works are often prone to re-authentication issues and increased handover delay. Therefore, to overcome the re-authentication process and provide users with better services, a novel handover authentication mechanism using deep learning (DHan_Auth) is proposed. Initially, the 5G data attack and normal data are collected, and the malicious and non-malicious users are classified using the Convolution Stacked long short term memory network model (Conv_SLSTM). To improve handover process and resist network attacks, only the non-malicious user data are authenticated through the key generation and the 5G Handover-Authentication and Key Agreement (5G_AKA) protocol. The process of encryption and decryption are performed using Extended Elliptic curve cryptography (Ex_ECC). When simulating with the PYTHON platform, performance such as handover latency, accuracy and precision are analyzed. The handover latency of the proposed model is 11.8 seconds to 200 nodes, while the classification accuracy in categorizing the malicious and non-malicious users reaches 98.98%.

783

Authors: Zaid S. Mohsen, Mohamed Jasim Mohamed

The inverted pendulum (IP) system, is a highly coupled, complex, nonlinear system in which the performance of the system is adversely affected by parameter uncertainty and outside disturbances. Therefore, these complications must be managed by the controllers created for such systems. The primary objective of this work is to develop four control structures, including integer order proportional integral derivative neural network controllers for inverted pendulums that deal with trajectory tracking issues. Proportional-integral-derivative neural network structure1 (PIDNNS1), proportional-integral-derivative neural network structure2 (PIDNNS2), proportional-integral-derivative neural network structure3 (PIDNNS3), and proportional-integral-derivative neural network structure4 (PIDNNS4) are the controller structures for inverted pendulum (IP) system . The ant colony optimization (ACO) is a metaheuristic optimization method that is offered to optimize. the controllers' settings while minimizing the cost function. The proposed controllers' resilience to outside disturbances and parameter uncertainty is also tested. The results using MATLAB code demonstrate that the PIDNNS4 controller, which best has a reduced cost function equal to (1.177494), (1.273627), (1.209761) for trajectory tracking, parameters uncertainty, and disturbances rejection for the inverted pendulum (IP) system. and the best controller for stabilization with a low-cost function is the PIDNNS1 controller (1.280839).

813

Authors: Merly Thomas, Bandu B. Meshram

In the modern era, communication through email is increased dramatically due to the cost-effectiveness, usage of contexts, application advertising, and so on. However, e-mails are considered a professional way of communication that helps both commercial and non-commercial organizations to share important documents, reports, and so on. Since e-mail acts as a global pathway, it attracts more intruders to create spam messages which result in storage consumption and virus attacks. To overcome these issues, an improved classification approach is introduced to classify spam emails from ordinary emails. The data is obtained from four benchmark datasets such as Enron, Lingspam, Spamassassin, TREC, and the pre-processing is performed using tokenization, lemmatization, and stemming. After this feature extraction is performed using Bag-of n-grams, latent dirichlet analysis (LDA), term frequency, and inverse document frequency (TF-IDF). Then the feature selection is performed using the proposed improved moth flame optimization (IMFO) algorithm and finally, the classification is performed using multi-class support vector machine (MSVM). The results obtained through experimental analysis show that the proposed IMFO-MSVM has achieved better accuracy of 98.68% whereas the existing semantic graph neural network (SGNN) and fuzzy rule based long short term memory (LSTM) have obtained accuracy of 97.87% and 97% respectively.

838

Authors: P. Oliver jayaprakash, M. Mohan Babu, B. Damodhara Reddy, S. Herald Lessly

Many countries face the traffic related issues regardless of the wide effects adopted for developing and improving the traffic flow. Depending on the drawbacks of existing traffic management systems like inefficient time management, absence of priority options and lack of dynamic and adaptive services, the need for the incorporation of intelligent concepts is mandatory. This facilitates the replacing of traditional approaches with an array of sensors and artificial intelligence (AI) through vehicle traffic control system. In this paper, the intelligent traffic system (ITS) is proposed on the basis of techniques such as internet of vehicles (IOV) and vehicular adhoc network (VANET). A road side unit (RSU) controller connected with the traffic control system acts as the key component of ITS. The on board unit (OBU) is fixed with each vehicle, which exchanges the real time information with RSU (road side unit) and other OBUs. Additionally, an adaptive traffic management algorithm (ATMA) is provided to govern the ITS for the generation of high performance outputs. Furthermore effectiveness of ATMA in the proposed work is proved by simulation using Python. Eventually, the ATMA is compared with fixed-time (FT) algorithm and traffic signal control algorithm (TSCA) under various traffic conditions. The findings depicts that the proposed ATMA-based ITS outperforms the current traffic management systems in terms of average waiting time (AWT) obtained as 277.5s with respect to number of vehicles and 282.14s with respect to number of roads at an intersection. Added to this average number of serviced vehicles is given by 1731 with respect to the number of roads.

863

Authors: Subashini K, Narmatha V

Phishing detection is a critical component of cyber security, aiming to safeguard users from malicious attempts to deceive and exploit. In this research, proposed an innovative approach that combines an adaptive threshold optimization technique with a learning-based graph convolutional network (GCN) and a scoring model to enhance phishing detection accuracy. The learning-based GCN is designed to analyze the intricate relationships within a graph structure, capturing nuanced dependencies between various entities such as email senders, recipients, and URLs. Leveraging this graph analysis, a scoring model assigns likelihood scores to instances being potential phishing attempts. Adaptive threshold optimization is a technique commonly used in feature selection to determine which features dynamically. In the context of phishing website detection, adaptive threshold optimization is used to select and prioritize the most informative features extracted from the websites. As a result of GCN, the model is able to differentiate between legitimate websites and phishing sites both locally and globally. In an effort to improve online security and safeguard users from phishing attacks, the proposed model is providing 99.3% accuracy in detecting phishing attacks. This enables a flexible and fine-tuned decision-making process, optimizing the trade-off between false positives and false negatives.

886

Authors: Rukmini S, Shridevi Soma

Cloud computing technology is widely used in many industries to build a large-scale data center. One of the main challenges in this technology is to minimize the energy utilization in data centres. The quality of service is remarkably obsessed with percentage of power utilization in cloud centres. Many methods have been explored to minimize the energy consumption and for effective Virtual Machine (VM) migration. But failed to attend the significant outcomes in migration and consolidation problems. Consolidation of VM into the minimal number of physical machines is considered one of the best solutions to consume minimum power. This research proposed an Energy Migration Optimized using Enhanced Pelican Remora (EMOEPR) for achieving the suitable placement of VM into minimum possible physical machine to diminish the number of overload hosts as much as possible. Also, it minimizes the energy consumption without violating service level agreement. The proposed method is simulated using Cloudsim Toolkit. The experimental procedure done on different workloads with various sizes of VM and PM, shows that does not affect service level agreement and outperforms compared with the existing techniques. The proposed method reduced 5% energy consumption, 22.62% of VM migrations, 8.62% of service level agreement violation when compared with the existing techniques.

909

Authors: Sanket Satish Mane, Ajit Kumar Santra

In this article, a novel optimization-assisted deep learning (OptiA-DL) strategy is proposed to effectively measure the temperature of visible heat sources. Initially, the image is acquired from the heat source and filtered using gelatin paper. The guided co-operative trilateral filter (GCTF) is used for image smoothening and HSV (Hue, Saturation, Value) or RGB (Red, Green, Blue) structure for colour transformation. Then, an improved Gaussian mixture model (IG-MM) is applied to segment the image into various portions. Accordingly, the compact split attention-based convolutional waderhunt autoencoder network (CSAtt-CWAN) is used for temperature prediction. In CSAtt-CWAN, the parameters are tuned using the waderhunt optimization algorithm (WHOA). The proposed OptiA-DL strategy is implemented in the Python platform through the candle flame dataset and assessed the performance in terms of different evaluation measures. Moreover, a proposed OptiA-DL’s performance is compared with existing methods to determine its efficiency in temperature measurement. The minimum mean square error (MSE) value obtained by the proposed OptiA-DL strategy is 0.076, which is lower than the existing methods for measuring the temperature of the visible heat source.

935

Authors: Shaikh Sohail Mohiyodin, Rajesh Maharudra Patil, M. S Nagaraj

The prevalence of renewable energy sources (RESs) and electric vehicles (EVs) is significant in modern electrical distribution systems (EDS). Although these technologies undoubtedly contribute to environmental improvement, they also pose considerable challenges for power system operators owing to their intermittent and unpredictable nature. To address these concerns, EDSs prioritise efficiency and reliability, which can be achieved through optimal network reconfiguration (ONR). The ONR problem addresses the challenge of incorporating varying levels of RESs and EVs into the system by considering the distribution losses and line loadability index (LLI). To overcome convergence issues in load flow and identify the optimal branches and tie lines for switching on/off, thus significantly improving network performance, a novel optimisation methodology that combines the northern goshawk optimisation algorithm (NGO) and LLI is proposed in this study. To enhance the search capabilities of the original NGO algorithm, a Levy Flight distribution and a new adaptive parameter were introduced, resulting in an improved version called Improved NGO (INGO). Simulations were conducted on a modified east delta network (EDN) of the unified Egyptian network (UEN), covering various scenarios. By having photovoltaic (PV) penetration, the network loss reduced to 698.46 kW from 723.85 kW, but by having EV penetration, the losses are raised to 848.15 kW, however, by having both PV and EV load penetration, the losses are reduced to 764.53 kW. However, by implementing ONR, the network loss was reduced to 739 kW. Furthermore, the computational efficiency of INGO is compared to that of the basic NGO, as well as other algorithms such as the stochastic fractal search (SFS), harmony search algorithm (HAS), and artificial rabbits algorithm (ARO). The results obtained from INGO surpassed those of all other algorithms in terms of the target function and computation time. Furthermore, the reduction in losses and enhanced loadability demonstrated the adaptability of the proposed methodology for real-time applications.

956

Authors: Barani Shaju, Valliammal Narayan

The prediction of the stock market offers information for the business to increase the profit in the share market. The prediction of a stock’s price is a challenging task because of the non-linearity, instability of time series information, and significant noise. In this research, an effective stock market prediction (SMP) is done by proposing the combination of long-short term memory (LSTM) and neural prophet (NP) namely the LSTMNP model. The interpretable features of NP are integrated with the LSTM to improve the prediction performances. On the other hand, the safety net incorporated in the LSTMNP is used to confirm the better prediction even if one of the models provides a lower performance. The main objective of this LSTMNP-SMP method is to improve the prediction with less errors. The LSTMNP-SMP method is analyzed using the parameters of mean absolute error (MAE), root-mean-square error (RMSE) and mean squared error (MSE). The existing approaches such as accelerated gradient-LSTM (AG-LSTM), support vector machine (SVM)-particle swarm optimization (PSO) and LSTM with artificial rabbits optimization (ARO) are used to compare the LSTMNP-SMP. The MSE of IBM for LSTMNP-SMP is 0.1110 which is less when compared to the AG-LSTM, SVM-PSO and LSTM-ARO.

981

Authors: Mohammad Khalaf Rahim Al-juaifari, Alih. Ali Athari

The prediction of human activities has garnered significant attention, owing to their relevance in diverse applications spanning healthcare, robotics, and user-computer interaction. This paper addresses the imperative need for a comprehensive multi-step prediction system tailored to accommodate these varied domains. The proposed framework comprises three key stages. Firstly, wavelet transform (WT) is employed for data pre-processing to eliminate noise and render the data amenable for time series analysis, the second phase of the framework involves the utilization of a hybrid model combining Long Short-Term Memory (LSTM) and Convolutional 1D (CONV1D) layers, denoted as LSTM-CONV1D designed to effectively address the complexities involved in extracting relevant features and tackling data imbalance issues. The LSTM component is employed for human activity classification based on sensor data, taking into account the sequential nature of activities. Concurrently, the CONV1D component is utilized for feature extraction. In the third phase, a method is introduced for predicting future activity levels and steps. This method incorporates a function that takes the output of the trained LSTM-CONV1D model as an input sequence. The performance of the proposed model is rigorously evaluated across four benchmark datasets based on sensor data: UCI-HAR (Accuracy: 99.03%), M-health (Accuracy: 93.21%), WISDM (Accuracy: 89.43%), and PAMAP2 (Accuracy: 86.85%). Notably, these results represent the state-of-the-art performance characterized by its simplicity attributed to the utilization of a single-layer LSTM and CONV1D. Furthermore, an additional evaluation is conducted on the UCI-HAR dataset with an accuracy of 98.92, wherein the LSTM time series function iteratively generates predictions for future time steps, following the same pre-processing steps.

1006

Authors: D. Shankar, G.Victo Sudha George, N. Kanya, S. Saraswathi

The intrusion detection system (IDS) plays an imperative role in defending the network from attacks. But, the IDS data is imbalanced, making the process complex for detecting the attacks accurately. According to these problems, this study proposes a network intrusion detection system based on an enhanced synthetic minority oversampling technique (SMOTE) and lightweight hybrid CAE (convolutional auto encoder)-ELM (extreme learning machine) model. Initially, the normalization of the original data is performed to avoid the influence of the maximum and minimum values. Secondly, an enhanced SMOTE is employed for solving the issues of the less detection rate of minority-attacks because of less training data. Here, the war strategy optimization (WSO) is incorporated with the SMOTE to design a novel WSO-based SMOTE technique for balancing the dataset. The three major steps utilized in WSO-based SMOTE are WSO based clustering, filtering and over-sampling. Then, the Information gain and fisher score based features extraction is employed for the reduction of computational complexity prior to the intrusion detection. Finally, the lightweight hybrid CAE-ELM model is executed for attack detection. CAEs are one of the most commonly used learning models because of their ability to construct a higher-level feature representation from the input data. Another model used in intrusion detection is the ELM, which provides an acceptable discrimination performance as well as a fast speed of learning. The performance of the proposed NIDS model is tested on the two benchmark datasets and achieved better accuracies of 99.21% and 99.15% on the UNSW-NB15 and CSE-CIC-IDS2018 datasets.