		FeedBack \| Contact Us \| Links \| Site Map

Welcome to Journal of Theoretical and Applied Information Technology

Submit Paper / Call for Papers

Journal receives papers in continuous flow and we will consider articles from a wide range of Information Technology disciplines encompassing the most basic research to the most innovative technologies. Please submit your papers electronically to our submission system at http://jatit.org/submit_paper.php in an MSWord, Pdf or compatible format so that they may be evaluated for publication in the upcoming issue. This journal uses a blinded review process; please remember to include all your personal identifiable information in the manuscript before submitting it for review, we will edit the necessary information at our side. Submissions to JATIT should be full research / review papers (properly indicated below main title).

Journal of Theoretical and Applied Information Technology
November 2025 | Vol. 103 No.22

Title:	ROBL-TSA AND SNAPSHOT ENSEMBLE CLASSIFIER FOR PAPAYA FRUIT DISEASE CLASSIFICATION
Author:	K. HARIKA , M.N. NACHAPPA
Abstract:	Papaya fruit is a healthy tropical fruit that is grown in many countries. But its growth and quality are affected by various diseases. Early and accurate disease detection is crucial to minimizing crop loss and the diseases appear in different shapes, sizes, and colours. In this study, developed a Snapshot Ensemble method to classify papaya fruit diseases more accurately. It integrates Histogram of Oriented Gradients (HOG) for feature extraction and Random Opposition-Based Learning with Tuna Swarm Algorithm (ROBL-TSA) to choose the best features for classification. The snapshot ensemble is utilized to improve generalization when reducing overfitting by using cyclic cosine annealing for adjusting learning rate in training. Our approach achieved 89.83% accuracy and 93.27% AUC on the papaya fruit disease dataset and performed better than traditional models like the Feed-forward Artificial Neural Network (FANN). This method offers a low-cost, effective solution for detecting papaya diseases.
Keywords:	Papaya Fruit Disease, Random Opposition-Based Learning, Snapshot Ensemble, Tuna Swarm Algorithm.
Source:	Journal of Theoretical and Applied Information Technology 30^th November 2025 -- Vol. 103. No. 22-- 2025
Full Text

Title:	PRIVACY IN IOT NETWORKS: A ZERO-KNOWLEDGE PROOF APPROACH FOR SHARDED BLOCKCHAIN ARCHITECTURES
Author:	YASSINE LKHALIDI, HAMZA MAJID ,HATIM KHARRAZ AROUSSI ,ACHRAF TIFERNINE
Abstract:	Internet of Things (IoT) deployments operate at the challenging intersection of constrained hardware, cryptographic security requirements, and distributed ledger scalability a problem space where existing blockchain IoT integrations fail due to unrealistic computational assumptions, monolithic trust models, or lack of privacy guarantees across shard boundaries. This paper bridges the gap between cryptographic theory and IoT reality by presenting a practical implementation of PLONK-based zero-knowledge authentication deployed on ESP32 hardware within a sharded blockchain architecture. We contribute: (1) A novel system architecture that integrates PLONK zk-SNARKs directly into the authentication layer of a sharded ledger, enabling IoT devices to prove identity without revealing credentials while maintaining cross-shard privacy; (2) A succinct cross-shard transaction protocol where zero-knowledge proofs accompany inter-shard transfers, ensuring atomic operations with minimal overhead addressing the previously unsolved problem of verifiable privacy across shard boundaries; (3) Practical validation on real ESP32 devices demonstrating sub-50ms verification times and 288-byte proof sizes, proving that modern zero-knowledge systems are deployable beyond theoretical constructs; (4) A game-theoretic security analysis showing negligible attack probabilities under realistic IoT network conditions. By validating PLONK on actual constrained hardware within a sharded architecture, we demonstrate that sophisticated cryptographic primitives can meet the stringent requirements of real-world IoT ecosystems. This work establishes that the integration of universal-setup SNARKs with sharded blockchains represents a practical path forward for securing billions of IoT devices without compromising privacy or scalability.
Keywords:	Elliptic Curve Cryptography, Zero-Knowledge Proofs, ZK-STARKs, PLONK, Shard Blockchain.
Source:	Journal of Theoretical and Applied Information Technology 30^th November 2025 -- Vol. 103. No. 22-- 2025
Full Text

Title:	DYNAMIC ADAPTIVEENHANCED ELLIPTIC CURVE CRYPTOGRAPHY FOR SECURE DATA ENCRYPTION AND DECRYPTION IN SMART CITIES
Author:	NALLURI BRAHMA NAIDU, GONDI LAKSHMEEWARI
Abstract:	As smart cities increasingly rely on interconnected systems, such as the Internet of Things (IoT) devices, sensors, and communication networks, the security of data conduction has developed a dangerousdisquiet. This research addresses securing data transmission in smart cities, where IoT devices and sensors handle sensitive data. It proposes Enhanced Elliptic Curve Cryptography (EECC) to provide a scalable, efficient, and resilient cryptographic framework ensuring confidentiality, integrity, and authentication. This model uses EECC to ensure protected data broadcast in smart cities, providing strong encryption, data confidentiality, integrity, and authentication. It addresses scalability, performance, and resilience to advanced security threats in interconnected systems. Contextual Adaptive Data Filtering (CADF) optimizes smart city security by dynamically adjusting encryption based on data sensitivity, network conditions, and threats. Dynamic Adaptive Elliptic Curve Encryption (DAECE) ensures secure, efficient communication in smart cities by adjusting encryption based on data sensitivity, network conditions, and threats.Enhanced ECC secures communication in IoT, smart grids, healthcare, blockchain, smart cities, mobile devices, and public safety. Homomorphic Encryption (HE-SDP) enables secure data analysis in smart cities, preserving privacy by processing encrypted data without revealing it. Findings show that ECC key sizes (100-500 bits) provide security levels of 75-250 bits, while RSA key sizes of 1000-15000 bits offer higher security at comparable levels, implemented in Python software. The future scope of EECC in smart cities includes further optimization for IoT devices, integration with emerging technologies like 5G and blockchain, and enhancing resilience against quantum computing threats for long-term security.
Keywords:	Elliptic Curve Cryptography, Encryption, Decryption, Smart Cities, Contextual Adaptive Data Filtering and Data Transmission.
Source:	Journal of Theoretical and Applied Information Technology 30^th November 2025 -- Vol. 103. No. 22-- 2025
Full Text

Title:	NON-LINEAR SMART MODEL SUPPORT VECTOR MACHINE (SVM) AS A HOAX NEWS CLASSIFIER
Author:	A M H PARDEDE, A FAUZI, Y MAULITA, PREDDY MARPAUNG, SUTRISNO ARIANTO PASARIBU
Abstract:	The urgency of this study lies in the increasing spread of hoax news in the digital era and its negative impact on society, particularly in relation to government policies such as budget efficiency and President Prabowo's free meal program. Hoaxes can trigger misinformation and undermine public trust in implemented policies. Therefore, an automated system capable of detecting hoax news with high accuracy is essential to counter the spread of misinformation. The objectives of this study are to: (1) develop a non-linear Support Vector Machine (SVM) model to improve the accuracy of hoax news classification; (2) identify the most influential features in hoax detection; (3) compare the performance of the non-linear SVM model with other machine learning methods; (4) design a web-based system or mobile application for automated hoax detection; and (5) apply the model to identify hoaxes related to public policy, thereby ensuring that the public receives valid and reliable information. The methodology adopts a Natural Language Processing (NLP)-based approach, consisting of several stages: data collection (hoax and valid news from various trusted sources), data preprocessing, feature extraction, implementation of the non-linear SVM model, model evaluation, and system development. The experimental results demonstrate that the proposed model achieves an accuracy of 93.6%, precision of 94.5%, recall of 92.9%, F1-score of 93.7%, and an AUC (Area Under the Curve) of 0.96. These results indicate that the model effectively distinguishes between hoax and non-hoax news. The high precision value suggests that the model is highly reliable in ensuring that news classified as hoax is indeed hoax. Meanwhile, the high recall value shows that the model is sufficiently sensitive in capturing most hoaxes, although there remain 40 hoax cases that were missed (false negatives). In conclusion, this study demonstrates that the non-linear SVM with an RBF kernel is a robust and reliable algorithm for classifying hoax and non-hoax news in the Indonesian language. Moreover, it successfully addresses the research questions posed at the outset, providing a significant contribution to automated hoax detection in the context of public policy and digital information integrity.
Keywords:	ews, Hoax, Classification, Linear, SVM.
Source:	Journal of Theoretical and Applied Information Technology 30^th November 2025 -- Vol. 103. No. 22-- 2025
Full Text

Title:	DEW COMPUTING WITH EDGE INTELLIGENCE FOR INDUSTRIAL AUTOMATION AND PREDICTIVE MAINTENANCE REAL-TIME ANOMALY DETECTION
Author:	AKASH GHOSH, ABHRANEEL DALUI, SATYENDR SINGH, SUNIL KUMAR SHARMA, LALBIHARI BARIK, JATINDERKUMAR R. SAINI, BIBHUTI BHUSAN DASH, UTPAL CHANDRA DE, SUDHANSU SHEKHAR PATRA
Abstract:	The increasing complexity of industrial automation systems, coupled with the pressing demand for real-time decision support, necessitates the deployment of efficient and decentralized computing paradigms. Edge computing (EC), operating at the periphery of the network, offers significant advantages by enabling localized data processing and reducing reliance on centralized cloud infrastructures. Building on this concept, this paper introduces a novel framework that integrates edge intelligence with dew computing (DC) to advance industrial automation and predictive maintenance. The proposed approach employs lightweight algorithms for real-time anomaly detection at dew nodes, enabling early identification of operational deviations in industrial equipment while maintaining minimal resource usage. Furthermore, causal inference models are incorporated to determine the root causes of equipment failures directly within the dew layer, thereby enhancing the precision of maintenance strategies and minimizing downtime. By leveraging localized computation, the framework effectively reduces latency, optimizes energy consumption, and enhances system reliability. Experimental evaluation demonstrates that the system achieves 96.3% accuracy in anomaly detection, correctly identifies root causes in 92.7% of cases, reduces average latency to 10.6 ms, and consumes only 2.4 W of power per dew node. A case study conducted in a smart manufacturing environment validates the practical benefits of the framework, highlighting improvements in anomaly detection and maintenance scheduling. The study also examines scalability and energy efficiency, underscoring the potential of the proposed system for deployment across diverse industrial settings.
Keywords:	Lightweight Anomaly Detection, Real-Time Industrial Systems, Fog Computing, Resource-Constrained Environments, Internet of Things (IoT), Predictive Analytics, Energy-Efficient Computing, Dew Computing
Source:	Journal of Theoretical and Applied Information Technology 30^th November 2025 -- Vol. 103. No. 22-- 2025
Full Text

Title:	A NOVEL FEATURE-SELECTION-GUIDED DEEP LEARNING FRAMEWORK INTEGRATING EXPLAINABLE AI FOR SCALABLE, DUAL-DIAGNOSIS PREDICTION OF DIABETES AND ADHD USING HETEROGENEOUS BIG HEALTH DATA
Author:	TANGUTURU SP MADHURI, DR. G. S RAGHAVENDRA
Abstract:	This study introduces a novel, feature-selection-guided deep learning framework designed to enable scalable and interpretable dual-diagnosis prediction of Diabetes and Attention Deficit Hyperactivity Disorder (ADHD) using heterogeneous health data. The model integrates structured clinical variables from electronic health records (EHRs), temporal patterns from wearable sensors, and behavioural features from psychological assessments. Advanced feature selection techniques including LASSO, Recursive Feature Elimination (RFE), and XGBoost were employed to extract 38 high-impact variables from over 400 inputs. These selected features feed into a multi-branch neural network comprising Convolutional Neural Networks (CNNs) for static inputs, Long Short-Term Memory (LSTM) networks for time-series data, and a Multilayer Perceptron (MLP) fusion layer for dual-output classification. The architecture achieved high accuracy for both Diabetes (93.2%) and ADHD (92.5%) with AUCs above 0.94, outperforming baseline models. Integrated explainable AI tools SHAP, LIME, and Grad-CAM enhanced interpretability, revealing key biomarkers such as HRV entropy, glucose variability, BMI, and impulsivity score. The model generalized well across demographic subgroups and demonstrated robust diagnostic capability even in comorbid cases. This work makes a distinctive contribution by embedding explainable AI directly within the deep learning training process, enabling transparent and adaptive model refinement. It also discovers shared physiological and behavioural biomarkers linking Diabetes and ADHD, offering new knowledge for comorbidity-aware healthcare analytics. This end-to-end framework provides a clinically transparent, technically rigorous solution for early, accurate, and interpretable multi-disease prediction, representing a significant step forward in comorbidity-aware healthcare AI systems.
Keywords:	Dual-diagnosis Prediction, Heterogeneous Health Data, Feature Selection, Deep Learning Architecture, Explainable Artificial Intelligence, Comorbidity Modelling
Source:	Journal of Theoretical and Applied Information Technology 30^th November 2025 -- Vol. 103. No. 22-- 2025
Full Text

Title:	PERFORMANCE EVALUATION OF WEIBULL-TYPE LIFETIME DISTRIBUTIONS BASED ON INFINITE FAILURE NHPP SOFTWARE RELIABILITY MODEL
Author:	SEUNG KYU PARK
Abstract:	This study focuses on evaluating the performance of NHPP-based software reliability models under infinite failure conditions by applying Weibull-type lifetime distributions, which are suitable for modeling time-dependent failure rates. Software failure time data were used for the analysis, and model parameters were estimated using the MLE approach. To quantitatively assess model performance, several evaluation criteria were employed: model efficiency measured by MSE and R², predictive accuracy assessed using the mean value function, failure occurrence rate using the intensity function, and system reliability based on the reliability function. The results demonstrate that the proposed Weibull-Extension model exhibits the highest performance among the models considered. These findings suggest that the Weibull-Extension model offers a more effective alternative for software reliability analysis and can be applied even in complex development environments. Therefore, this study contributes to the field by newly identifying the reliability performance of Weibull-type models, which have been underexplored in previous research, and is expected to support developers in early-stage failure rate estimation during the software development lifecycle.
Keywords:	Extended-Weibull, Infinite-Failure NHPP, Musa-Okumoto, Weibull, Weibull-Extension, Weibull-Type.
Source:	Journal of Theoretical and Applied Information Technology 30^th November 2025 -- Vol. 103. No. 22-- 2025
Full Text

Title:	BERT-GCN MULTI-TASK MODEL FOR DISASTER DETECTION, CLASSIFICATION, AND HELP IDENTIFICATION IN TWEETS
Author:	BASUDEV NATH, DEEPAK SAHOO, SATYENDR SINGH, SUNIL KUMAR SHARMA, SUDHANSU SHEKHAR PATRA
Abstract:	X, which used to be called Twitter, has become an important way to share information during natural disasters in this age of real-time digital contact. It is still very hard to get useful information from the huge amounts of social media data that are available. Our study suggests a multi-stage integrated deep learning models that uses both contextual and relational features to effectively sort and analyze tweets about disasters. In the first task, our model uses binary classification to figure out with 96.58% accuracy if a tweet is disaster related or not. A multi-class categorization of tweets into three categories—non-disaster, flood, and earthquake—is performed in the second task. The accuracy of this classification is 92.65%, 96.30%, and 95.48%, respectively. The third and most focused task is to identify specific support needs in earthquake-related tweets, such as calls for food and medical help, with an accuracy of 95.07%. Graph Convolutional Networks (GCNs) are used to learn the relationships between words, and BERT (Bidirectional Encoder Representations from Transformers) is used to get deep contextual meanings. The combined embeddings are passed into a fully connected neural network (FCNN) for classification. Experimental findings show that our hybrid approach not only works well across all tasks but also has practical relevance in improving real-time catastrophe response and focused humanitarian relief distribution.
Keywords:	Tweeter Analysis, BERT, GCN, NLP, FCNN, Tweeter Help Categorization
Source:	Journal of Theoretical and Applied Information Technology 30^th November 2025 -- Vol. 103. No. 22-- 2025
Full Text

Title:	SCALABLE AND SECURE MULTI-USER AUDIO COMMUNICATION IN CLOUD-BASED SYSTEMS USING BLOCKCHAIN AND ENCRYPTION
Author:	DHANARAJU MURALA, Dr.K.THAMMI REDDY
Abstract:	In recent advancements, the integration of encrypted data frameworks with blockchain systems like the Hierarchical-Policy Identity-Centric Cryptographic Algorithm (HP-ICCA) has captured the scholarly interest due to its potential to ensure comprehensive security assessment and transaction goavousibility in data exchange domains. The prevalent blockchain-centric HP-ICCA paradigms, however, often leave cryptographic keys under the aegis of a singular centralized authority, resulting in intensive computational demands, elevated transactional expenditures, and limitations in scalability within a decentralized schema. To mitigate these issues, our study introduces an enhanced strategy for employing a distributed encryption mechanism (DESM) WITH zero-knowledge protocol enhancements. In a blockchain network, expansion into areas addressing Ciphertext-Policy Attribute-Based Encryption (CP-ABE) is indispensable. Common impediments include centralized issuance dependent on variable ζ to define authorities .Ciphertext Policy Attribute-Based Encryption (CP-ABE) models are prevalent in cloud- sharing scenarios but suffer from privacy issues in access policies, user or attribute irrevocability, key escrow dilemmas, and trust constraints. To enhance CP-ABE’s practicality while maintaining ζ (key management security), distributed models (Ω) along with zero-knowledge models λ for encryption are explored. By deploying proxy re-encryption nodes (Ρ), which ensure multi-party management and dissemination of master keys (ΜΚ), autonomy from central authorities is achieved, contributing to the establishment of comprehensive reencryption proofs ρ . Our proposal contemplates an economic incentive model thereby utilizing staking mechanics (Σ) to devise an equilibrium between security and reward distribution. It includes mathematical optimizations for latency reduction to achieve optimal performance, e.g., in scenarios exhibiting concurrent transaction processing of up to 100 occurrences instantaneously, and a reduction in average gas consumption γ greater than 97%.
Keywords:	Audio Based Block Chain, Cloud Computing, Multiuser Security.
Source:	Journal of Theoretical and Applied Information Technology 30^th November 2025 -- Vol. 103. No. 22-- 2025
Full Text

Title:	CUSTOM REGRESSION-BASED ARCHITECTURES FOR REAL-TIME TRAJECTORY PREDICTION USING SPATIO-TEMPORAL DATA
Author:	PRAVEEN KUMAR .V .S , SAJIMON ABRAHAM , SIJO THOMAS, SABU AUGUSTINE
Abstract:	Trajectory prediction plays a critical role in applications such as autonomous navigation, human motion forecasting, and robotic path planning. However, despite its importance, the scarcity of clean, large-scale datasets and the high bandwidth cost of continuous polling limit practical deployment in real-world intelligent systems. This study proposes a comparative evaluation framework using a hybrid dataset that integrates real and simulated user travel trajectories. The performance of seven regression models—LSTM, Bidirectional LSTM (Bi-LSTM), GRU, 1D CNN, Transformer, SVM, and XGBoost—is systematically analyzed using key metrics including Mean Squared Error (MSE), Mean Absolute Error (MAE), Root Mean Squared Error (RMSE), R-squared (R²), and Symmetric MAPE (SMAPE). The goal of this study is to benchmark these models on hybrid trajectory datasets and to test the hypothesis that sequence-aware deep models (Bi-LSTM, GRU) outperform traditional learners (SVM, XGBoost), with SMAPE providing clearer insights for low-value targets. Among these, the Bi-LSTM model demonstrates superior performance, particularly in terms of MAE and SMAPE. Additionally, the study explores hyperparameter tuning techniques, statistical validation, and the challenges of combining real and synthetic data, highlighting the effectiveness of SMAPE in low-value target scenarios. The findings not only support the use of Bi-LSTM for enhanced trajectory prediction accuracy in hybrid data environments but also underline the broader potential of predictive intelligence in reducing data transmission overhead and improving the efficiency of real-time mobility systems. These findings have direct implications for smart mobility and surveillance applications, where reducing polling frequency without compromising accuracy is essential. This study directly addresses the critical challenge of accurate real-time trajectory prediction under conditions of data scarcity by integrating both real and simulated datasets. The central goal is to design and benchmark regression-based architectures for short-term forecasting and to test the hypothesis that prediction accuracy can be significantly enhanced by incorporating temporal dependencies within spatio-temporal data.
Keywords:	Trajectory Prediction, Bidirectional LSTM (Bi-LSTM), Real And Simulated Data, Regression Models, Performance Evaluation Metrics, SMAPE
Source:	Journal of Theoretical and Applied Information Technology 30^th November 2025 -- Vol. 103. No. 22-- 2025
Full Text

Title:	FACTORS THAT INFLUENCE CUSTOMER CONTINUANCE USAGE OF DIGITAL STREAMING VIDIO SERVICES
Author:	FERDY RENHART SITORUS, ELFINDAH PRINCES
Abstract:	The purpose of this study is to identify the factors that influence customers in the continuance usage of the Vidio digital streaming service using a quantitative research method. This study employs data analysis techniques with PLS-SEM to examine the relationships among latent variables. Data were collected through questionnaires distributed to 393 respondents, of which 385 had active subscriptions to the Vidio application. The results of the study show that eight hypotheses were accepted. The variable perceived usefulness has a significant effect on customer satisfaction. Perceived usefulness also significantly influences continuance usage. Then, the variable confirmation significantly affects perceived usefulness. Confirmation also has a significant effect on customer satisfaction. The variables content quality, price, and personalization significantly influence customer satisfaction, which in turn affects continuance usage. Finally, customer satisfaction has a significant influence on continuance usage.
Keywords:	Video on Demand, Digital Streaming Vidio Services, Customer Satisfaction, PLS-SEM, Continuance Usage
Source:	Journal of Theoretical and Applied Information Technology 30^th November 2025 -- Vol. 103. No. 22-- 2025
Full Text

Title:	AI-POWERED SMART CITY EVACUATION PLANNING: TRAFFIC CONGESTION PREDICTION USING DEEP BELIEF NETWORKS
Author:	GEETHA PAWAR1, B SARITHA, RAVI UYYALA, VIJAYA CHANDRA JADALA, RAYAVARAPU VEERANJANEYULU, RUDRAMANI BHUTIA, NIDAMANURU SRINIVASA RAO
Abstract:	This study aims to evaluate the feasibility and efficacy of using Artificial Intelligence (AI) Deep Learning in smart city contexts. A traffic flow prediction model is developed utilizing the Deep Belief Network (DBN) algorithm. The designated road segment and its historical traffic flow data in Tianjin have been gathered and pre-processed. Subsequently, many Restricted Boltzmann Machines (RBMs) are aggregated to construct a Deep Belief Network (DBN), which is trained as a generative model. The performance is ultimately assessed using a simulation exercise. The suggested algorithm model is compared with the Neuro Fuzzy C-Means (FCM) model, Deep Learning Architecture (DLA), and Convolutional Neural Network (CNN) model. The findings indicate that the Root Mean Square Error (RMSE), Mean Absolute Error (MAE), and Mean Absolute Percentage Error (MAPE) of the suggested algorithm model are 4.42%, 6.21%, and 8.03%, respectively. Its predictive accuracy surpasses that of the other three algorithms considerably. Furthermore, the algorithm can efficiently mitigate the proliferation of congestion in the smart city, facilitating prompt alleviation of traffic bottlenecks. The developed Deep Learning-based traffic flow prediction model demonstrates great precision in forecasting and effective traffic congestion mitigation, offering valuable experimental insights for future smart city development.
Keywords:	Smart cities, Deep Learning, Traffic flow prediction, Artificial Intelligence, Deep Belief Network
Source:	Journal of Theoretical and Applied Information Technology 30^th November 2025 -- Vol. 103. No. 22-- 2025
Full Text

Title:	REDISTRIBUTION OF SOCIAL RESOURCES IN THE MODERN ERA: NEW CHALLENGES FOR THE STATE FINANCE SYSTEM
Author:	OLEKSANDR BULAVYNETS, VOLODYMYR VALIHURA, IRYNA SYDOR, KATERYNA KRYSOVATA, VIKTORIA BELIAVTSEVA
Abstract:	Relevance. The digital transformation of public finances and social transfers is becoming particularly important in the context of rapid technological change and growing public demand for transparency, efficiency, and accessibility of public services. Given that the introduction of digital tools allows for the optimization of budget resource management, improves the efficiency of social support, and promotes the sustainable development of social systems, it is necessary to take these processes into account in order to modernize public administration and ensure social justice. Aim. The aim of this study is to determine the impact of social transfers on the efficiency of public finances in the context of digital transformation and to develop recommendations for their optimization. Methods. The study used methods of literary source synthesis, statistical data analysis, systematization, and generalization to study the impact of social transfers on the effectiveness of public finances in the context of digital transformation, which revealed uneven social spending in EU countries and the importance of digitalization for improving the effectiveness of social policy. Results. The analysis showed that social transfers significantly increase the financial capacity of local communities, but the level of social spending varies significantly between EU countries, and the introduction of digital technologies improves the transparency and management of social benefits. Conclusions. Based on the analysis, it was determined that in order to improve the effectiveness of social policy, it is necessary to combine social transfers with the development of digital infrastructure and the introduction of innovative management practices.
Keywords:	Social Transfers, Public Finance, Digitalization, Digital Technologies, Innovative Digital Solutions, Social Benefit Management, Social Policy Effectiveness, Transformations, Budget, Financial Analysis, Fiscal Policy.
Source:	Journal of Theoretical and Applied Information Technology 30^th November 2025 -- Vol. 103. No. 22-- 2025
Full Text

Title:	AN ADAPTIVE GRAPH NEURAL NETWORKS WITH LANDSCAPE-AWARE PARTICLE SWARM OPTIMIZATION FOR INTELLIGENT MEDICAL INSURANCE FRAUD DETECTION
Author:	MR. V VINAY KUMAR, M V V A L SUNITHA, PALAMAKULA RAMESH BABU, A ARUNA KUMARI ARAJU ANITHA, BPRAVEENA MANDAPATI
Abstract:	This paper introduces a novel framework for detecting fraudulent medical insurance claims using Adaptive Graph Neural Networks (AGNN) optimized by Landscape-Aware Particle Swarm Optimization (LAPSO). By modeling healthcare data—including patients, providers, and procedures—as a graph, AGNN captures complex relational patterns indicative of fraud. Dynamic attention mechanisms help highlight critical relationships while LAPSO tunes key hyperparameters to improve generalization. Experimental results on real-world datasets demonstrate the proposed model’s superiority in terms of accuracy (91%), precision (89%), recall (88%), and F1-score (88%) over traditional machine learning and deep learning models. The results confirm the efficacy and interpretability of our framework for practical fraud detection applications. The AGNN component employs dynamic attention mechanisms to selectively prioritize significant relationships during message passing, thereby enhancing the ability of the model model’s ability to detect subtle and coordinated fraud schemes. To further improve detection accuracy and generalization, LAPSO fine-tunes critical hyperparameters of the AGNN architecture. By leveraging both global and local search capabilities, LAPSO accelerates convergence toward optimal configurations while adapting to the model’s performance landscape.The proposed method is evaluated on a real-world medical claim’s dataset, demonstrating superior performance associated with conventional deep learning and graph-based models among key metrics, including accuracy, precision, recall and F1-score. Moreover, the framework exhibits strong generalization capability, effectively identifying both known and previously unseen fraudulent behaviors. The integration of graph learning and evolutionary optimization offers a scalable and interpretable solution for healthcare providers and insurance companies aiming to mitigate fraud risks[15]. This study contributes to the progression of intelligent fraud detection systems and opens new directions for the application of adaptive graph learning in health. The AGNN component employs dynamic attention mechanisms to selectively prioritize significant relationships during message passing, improving the model’s capability to notice subtle and coordinated fraud schemes. To further improve detection accuracy and generalization, LAPSO fine-tunes critical hyperparameters of the AGNN architecture. By leveraging both global and local search capabilities, LAPSO accelerates convergence toward optimal configurations while adapting to the model’s performance landscape.
Keywords:	Adaptive Graph Neural Networks(AGNNs),LAPSO,Deep Learning ,Medical Insurance,Health care analytics etc.,
Source:	Journal of Theoretical and Applied Information Technology 30^th November 2025 -- Vol. 103. No. 22-- 2025
Full Text

Title:	A HYBRID CYBERSECURITY FRAMEWORK FOR SMART EV CLOUD SYSTEMS USING FUZZY LOGIC, MACHINE LEARNING, AND BLOCKCHAIN
Author:	PANTHANGI VENKATESWARA RAO, SK. MD. SHAREEF, DEEPIKA VODNALA , SIRISHA NARKEDAMILLI , ASHISH B. JIRAPURE , P. LAKSHMI PRASANNA , P. S. SUBHASHINI PEDALANKA
Abstract:	The increasing need for communication within and between electric vehicles can create serious challenges for infrastructure. This paper focuses on protecting electric cars from cyberattacks by introducing a secure and intelligent framework. We propose a new cybersecurity method that combines blockchain technology with smart cloud computing and fuzzy machine learning, specifically designed for electric vehicle systems. To handle data from vehicles, the model uses a cloud system integrated with the smart grid, while the fuzzy adversarial Q-stochastic (FAQS) model detects and analyzes suspicious activities. Data is protected through encryption and decryption, based on role-based access control, which ensures only authorized us-ers can access information according to their responsibilities. The proposed system was tested using differ-ent cybersecurity datasets and evaluated on performance measures such as security rate, error (RMSE), quality of service (QoS), scalability, and energy efficiency.
Keywords:	Electric Vehicle, Smart Cloud Computing, Cyber Security Analysis, Fuzzy Machine Learning, Blockchain Model
Source:	Journal of Theoretical and Applied Information Technology 30^th November 2025 -- Vol. 103. No. 22-- 2025
Full Text

Title:	CONTEXT-AWARE PRIORITIZATION OF SOFTWARE REQUIREMENTS FOR SMALL-SCALE PROJECTS USING MODAL VERBS, SEMANTIC EMBEDDINGS, AND GRAPH-BASED RANKING
Author:	SUCHETHA VIJAYAKUMAR , SURESHA D
Abstract:	Effective prioritization of software requirements is critical to the success of software projects. Traditional methods such as MoSCoW often lack the ability to semantically distinguish and set a requirements based on context. This paper presents a unified approach that first leverages context-aware and semantic ordering of requirements within each priority group of High, Medium and Low priority requirements. The ordering is done by ten different approaches belonging to various categories and the results of the same are compared. Unlike traditional MoSCoW-based or purely semantic approaches, our method integrates linguistic modality, contextual semantics, and graph-based learning in a unified ranking framework, enabling more nuanced, accurate requirement prioritization. Extensive experiments on a real-world dataset demonstrate that ModalGraphIntegration, a hybrid method involving fusion of semantic, linguistic, and graph-based signals achieves superior performance across Exact Match Accuracy, Top-K Accuracy, MRR, and NDCG.
Keywords:	Functional Requirement Prioritization, Modal Verbs, Natural Language Processing, Requirement Engineering, Software Development, Automated Prioritization.
Source:	Journal of Theoretical and Applied Information Technology 30^th November 2025 -- Vol. 103. No. 22-- 2025
Full Text

Title:	ANALYSIS OF A HARMONIZED OPTIMIZATION-DRIVEN CROSSOVER PERCEPTRON NETWORK FOR HEART DISEASE PREDICTION
Author:	M. K. ARIF, KALAIVANI KATHIRVELU, A.YASIR
Abstract:	Heart disease prediction requires accurate and efficient models capable of handling heterogeneous medical data, mitigating noise distortions, and optimizing feature selection for enhanced classification performance. To address these challenges, we propose the Harmonized Optimization-Driven Crossover Perceptron Network (HOC-Perceptron), an integrated framework comprising Advanced Normalization-Noise Filtering (ANNF) for robust data preprocessing, Greylag Goose Optimization (GGO) for feature selection, and Crossover Arithmetic-Optimized Multi-Layer Perceptron (CAO-MLP) for classification. ANNF improves signal quality by 27.8%, reducing noise-induced distortions while preserving diagnostic markers. GGO enhances feature selection efficiency by 32.5%, ensuring optimal subset selection and reducing computational complexity. CAO-MLP further boosts classification accuracy by 8.9%, achieving an overall F1-score of 96.4%, outperforming baseline models in terms of convergence speed and generalizability. The proposed HOC-Perceptron framework significantly enhances heart disease prediction reliability, offering a computationally efficient and clinically interpretable solution for early diagnosis and risk assessment.
Keywords:	Heart disease prediction, ECG signals, Greylag Goose Optimization (GGO), Feature Selection, Crossover Multi-Layer Perceptron (CMLP) Arithmetic Optimization Algorithm (AOA)
Source:	Journal of Theoretical and Applied Information Technology 30^th November 2025 -- Vol. 103. No. 22-- 2025
Full Text

Title:	AN INTELLIGENT HYBRID NEURAL-SWARM ARCHITECTURE FOR ADAPTIVE AND ENERGY-EFFICIENT LOAD BALANCING IN CLOUD COMPUTING
Author:	AHMED KHALIFA, HELAL A. SULEMAN, MOHAMED MARIE
Abstract:	Cloud computing delivers flexible, transparent, and interactive access to shared resources over the Internet, freeing consumers from device management concerns and providing seamless data and computational access anytime, anywhere. The pervasive adoption of smart devices has necessitated the rapid expansion of cloud computing, placing load balancing at the forefront of resource management challenges. Effective load balancing allows for equitable distribution of tasks among multiple devices, which is vital for maximizing performance and resource utilization in cloud environments. This paper introduces a hybrid Model that leverages Deep Neural Networks (DNNs), Long Short-Term Memory (LSTM), and Particle Swarm Optimization (PSO) to address prevailing research gaps in cloud load balancing optimization. Our proposed approach focuses on resource provisioning and dynamic load balancing among virtual machines (VMs), aiming to achieve optimal utilization and uniform workload distribution. The model is trained and evaluated on a complex cloud load balancing dataset, demonstrating its capacity to distinguish optimal load balancing scenarios under real-world conditions. Experimental results highlight the efficacy of the proposed system, achieving a test accuracy of approximately 92% and a test AUC of around 0.98. However, while validation AUC exceeded 0.99, a notable gap in test AUC (~0.49) indicates Complexity in model tuning, which is discussed in detail. Overall, the hybrid DNN-LSTM-PSO model presents a robust and scalable solution for load balancing in cloud computing, effectively integrating deep learning’s hierarchical feature extraction and sequence modeling with advanced optimization strategies for improved cloud service performance.
Keywords:	Cloud Computing, Load Balancing, Hybrid Load Balancing Algorithms, Deep Neural Networks, Long Short-Term Memory, Particle Swarm Optimization
Source:	Journal of Theoretical and Applied Information Technology 30^th November 2025 -- Vol. 103. No. 22-- 2025
Full Text

Title:	STRATEGIC MANAGEMENT OF OMNICHANNEL MARKETING WITH THE USE OF GENERATIVE ARTIFICIAL INTELLIGENCE
Author:	SVITLANA KOVALCHUK, TETIANA CHUNIKHINA, VADIM H. GRIGORYAN, VALENTYNA SHEVCHENKO, ULIANA KOROLOVA
Abstract:	The relevance of the issue under research is determined by the rapid introduction of generative artificial intelligence (GenAI) into omnichannel marketing strategies. This radically changes approaches to personalization, communication, and strategic management in the digital economy. There is a growing need for quantitative analysis of the impact of such technologies on marketing efficiency, which justifies the need for this study. Particular attention is paid to personalization, communication automation, and data integration. The aim of the study is to determine the relationship between the level of implementation of GenAI and marketing efficiency. The problem is the lack of quantitative assessments of the impact of artificial intelligence (AI) on omnichannel management indicators. The analysis covers 10 companies from 10 countries for 2022-2024. An econometric model with six variables was applied: GAI, AdSpend, Data Integration, CSAT, OMDEPTH, and BRAND. The highest OM_EFF values in 2024 were recorded at Bosideng (98.3), Woolworths (97.0), and Zara (96.9). Companies with high GenAI have seen faster OM_EFF growth and market adaptation. Woolworths’ GenAI increased from 0.58 to 0.77, OM_EFF from 85.7 to 97.0. Hudson’s Bay showed the smallest changes because of limited digitalization and weak AI development. The study found that a high level of integration of GenAI is directly related to the increase in omnichannel marketing efficiency (OM_EFF). Bosideng achieved the highest OM_EFF – 98.3 in 2024 – with the maximum value of the GenAI Index (0.83) and the active use of personalized content. The obtained data confirm that GenAI enhances the effectiveness of marketing strategies through automation, deep data integration and multi-channel interaction with customers. The article provides practical recommendations for increasing OM_EFF. The study has applied significance for retail trade strategies. Further research may include other industries and regions.
Keywords:	Omnichannel Marketing, Generative AI, Efficiency, Strategic Management, Customer Personalization, Digital Transformation, Econometric Model
Source:	Journal of Theoretical and Applied Information Technology 30^th November 2025 -- Vol. 103. No. 22-- 2025
Full Text

Title:	A CROSS-LAYER ROUTING FRAMEWORK FOR ENERGY-EFFICIENT AND LOW-LATENCY IOT COMMUNICATION: SIMULATION AND EXPERIMENTAL VALIDATION
Author:	YUVA KRISHNA ALURI , S TAMILSELVAN
Abstract:	The rapid expansion of the Internet of Things (IoT) has created unprecedented challenges in providing efficient, low-latency, and energy-aware communication among resource-constrained devices. Traditional routing approaches are designed in isolation from each other at different layers of the OSI model and therefore do not consider the interdependencies between as an example of the physical, MAC, and network layers. In this work, we present a Cross-Layer Routing Framework (CLRF), which dynamically integrates metrics of residual energy (RE), link quality (LQ), and latency (E2E latency to each node) to provide efficient routing in heterogeneous IoT environments. The validity of the framework is established by extensive simulation studies carried out in NS-3 and experimental deployment as a testbed using an ESP32-Raspberry Pi combination. The simulation and experimental results show that CLRF achieves up to 25% improved energy efficiency, 40% improved average end-to-end latency, and 12% improved packet delivery ratio (PDR) compared to traditional baseline protocols such as LEACH and RPL. The statistical validity of the improvements is confirmed through Wilcoxon signed-rank test with 30 independent runs (p < 0.05) and bootstrap confidence intervals further reinforce the reliability of the experimental findings. The experimental results show similar behavior to the simulation results and it indicates the robustness and scalability of the framework. This research proposes a unified framework addressing dual problems of energy efficiency and delay which allows for sustainable and real-time communication in IoT communications.
Keywords:	Cross-Layer Design, Iot Communication, Energy Efficiency, Low-Latency Routing, NS-3 Simulation, Experimental Validation
Source:	Journal of Theoretical and Applied Information Technology 30^th November 2025 -- Vol. 103. No. 22-- 2025
Full Text

Title:	STREAMLINING REQUIREMENTS ANALYSIS USING LARGE LANGUAGE MODELS AND USER INTERFACE AUTOMATION
Author:	MAMATHA TALAKOT, BALARAM AMAGOTH , SUDHA RANI CHIKKALWAR, SUDHAKAR JANGILI, ANGOTU NAGESWARA RAO, RAVI MOGILI
Abstract:	Requirements analysis, a critical phase in software development, often grapples with the challenge of extracting structured user stories from unstructured data sources such as text descriptions, UI mockups, and informal communication. This paper presents a novel approach leveraging Large Language Models (LLMs) and prompt engineering to automate the generation of well-structured user stories. By employing carefully crafted prompts, LLMs can effectively analyze unstructured data, identify key requirements elements (user roles, goals, benefits), and translate them into actionable user stories. Proposed work evaluate the effectiveness of this LLM-driven approach through a comparative study with manually created user stories, assessing accuracy, completeness, and clarity. Preliminary results demonstrate the potential of LLMs to streamline requirements analysis, improve the quality of user stories, and ultimately contribute to the development of software solutions that better align with user needs. However, challenges such as hallucinations and inconsistencies in LLM-generated outputs warrant further investigation and refinement. This research provides valuable insights into the feasibility and limitations of using LLMs for automating user story generation, paving the way for more efficient and effective requirements engineering practices.
Keywords:	Software Requirements, User Stories, Prompt Engineering, Requirement Analysis, Speech-Text Engine, LLMS
Source:	Journal of Theoretical and Applied Information Technology 30^th November 2025 -- Vol. 103. No. 22-- 2025
Full Text

Title:	DSS QUERY OPTIMIZATION USING ENTROPYBASED RESTRICTED GENETIC APPROACH ANDPARALLEL PROCESSING
Author:	K. SWARUPA RANI,B. LAKSHMI,Dr. YARLAGADDA ANURADHA, Dr. KOLLURU SURESH BABU,P. VENKATESWARLU REDDY,CH. SABITHA, DIVVELA.SRNIVASA RAO
Abstract:	Any distributed database system's query optimization challenge is exciting and continues to attract a lot of interest. Recent years have seen the application of a number of heuristics that suggest novel strategies for enhancing a query's viability. One of the most widely used heuristics for optimization issues is the Genetic Algorithm (GA). The search for a better solution is still ongoing. This work proposed the Entropy-based Restricted Genetic Approach (ERGA), a novel query optimizer approach. Massive amounts of data (in gigabytes, petabytes, or even more) are processed using DSS queries. As a result, these queries are unlikely to be mission impart antin terms of "Response Time." However, a major concern is the amount of system resources required to execute the query. As a result, Total Costs—the sum of input, output, processing, and communication costs—are used to optimize DSS enquiries. The overall cost of DSS queries can be decreased by increasing the replication factor in the distributed database. It was determined that ERGA more effectively met the low time complexity (Runtime) and high quality (Total Costs) objectives for distributed DSS queries, which are otherwise at odds with one another. Intra-Site Parallelism is proposed to reduce ERGA's performance. Two or more processors may be present at a site in intra-site parallelism.The utilization of two cores or processors over a site has been shown to significantly enhance ERGA quality by up to 6.5%. Additionally, several of the standard metrics of descriptive statistics were used to statistically analyze ERGA. Using ERGA, the best solution outperformed the worst by 89%.
Keywords:	Query Optimization, DSS, Genetic Algorithm, ERGA, Parallel Processing
Source:	Journal of Theoretical and Applied Information Technology 30^th November 2025 -- Vol. 103. No. 22-- 2025
Full Text

Title:	HOW CAN GRAPH NEURAL NETWORKS AND ATTENTION MECHANISMS IMPROVE HUMAN ACTIVITY RECOGNITION? A MULTIMODAL DEEP LEARNING FRAMEWORK
Author:	SAURABH GUPTA, RAJENDRA PRASAD MAHAPATRA
Abstract:	HAR is a widely studied area that works on recognizing human actions from information captured by different sensors such as cameras and inertial sensors. The performance of HAR has been greatly enhanced thanks to recent developments in deep learning, mainly because of convolutional and recurrent neural networks. It provides a thorough overview of HAR techniques between 2020 and 2025, focusing on models that blend data from RGB images, depth images and skeleton joints together. Our design combines the benefits of Xception and EfficientNet for feature extraction, along with skeleton-based features, to make the recognition more accurate and solid. Tests conducted on recognized UTD-MHAD, HMDB51 and UCF101 benchmarks prove that the model outperforms other methods, surpassing 92.79% accuracy. Furthermore, the paper addresses the issues brought by dataset limits, complex computing requirements and difficulties in adapting the models to new applications and proposes promising paths for advancements in HAR.
Keywords:	Human Activity Recognition (HAR), Deep Learning, Multi-Modal Data Fusion, Convolutional Neural Networks (CNN), Recurrent Neural Networks (RNN), Skeleton Joint Analysis, Benchmark Datasets, Hybrid Models, UTD-MHAD.
Source:	Journal of Theoretical and Applied Information Technology 30^th November 2025 -- Vol. 103. No. 22-- 2025
Full Text

Title:	MULTIPLE SCLEROSIS MR IMAGES DENOISING MODEL FOR ACCURATE CLASSIFICATION USING CONVOLUTION NEURAL NETWORK
Author:	KALPANA SANJAY PAWASE, LUBNA TARANUM M P, VAIBHAV VASUDEVRAO GIJARE, SHABANAM KHALID SHIKALGAR, ARIFA JAVID SHIKALGAR
Abstract:	Multiple Sclerosis (MS) is a chronic neurological disease characterized by demyelination in the central nervous system, where Magnetic Resonance Imaging (MRI) plays a vital role in lesion detection. However, accurate identification of MS lesions remains a challenge due to the presence of various noise types such as Gaussian, speckle, and salt-and-pepper noise, which reduce image clarity and distort lesion boundaries. Existing denoising and segmentation approaches, including traditional filters and Convolution Neural Networks (CNN) based models, have shown limitations, filters often compromise edge preservation, while many deep learning techniques require extensive pre- and post-processing, leading to reduced efficiency and inconsistency in lesion classification. This gap in the literature highlights the need for an approach that can simultaneously suppress noise and retain essential lesion details for reliable segmentation. To address this, the study proposes a Hybrid Convolutional Neural Network with Pixel Batch Normalization (HCNN-PBN) that integrates denoising and segmentation in a unified framework. The novelty of this model lies in its ability to preserve critical structural features during the denoising process through pixel batch normalization, thereby improving lesion classification accuracy compared to conventional CNN-based denoising methods. The research introduces new knowledge by demonstrating that denoising and segmentation can be optimized together in a hybrid CNN framework without significant loss of detail, reducing computational complexity while enhancing diagnostic reliability. Experimental results validate that the proposed HCNN-PBN model outperforms existing CNN-based approaches in terms of denoising accuracy, segmentation accuracy, filtering efficiency, and processing time, thereby offering a more effective solution for automated MS lesion detection in MRI images.
Keywords:	Multiple Sclerosis, Magnetic Resonance Imaging, Convolution Neural Networks, Denoising, Segmentation. Pixel Normalization.
Source:	Journal of Theoretical and Applied Information Technology 30^th November 2025 -- Vol. 103. No. 22-- 2025
Full Text

Title:	THE USE OF ARTIFICIAL INTELLIGENCE SYSTEMS TO SUPPORT FORENSIC ACTIONS IN COMPUTER CRIMES INVESTIGATIONS
Author:	VIACHESLAV KULIUSH, YULIIA CHORNOUS, OLEKSII KHARKEVYCH, OKSANA VASYLOVA, VITALII MATSAK
Abstract:	The relevance of the study is determined by the increasing number of computer crimes and the need for a normatively permissible, explainable, and integrated implementation of artificial intelligence (AI) systems in digital forensics. The aim of the research is to develop an optimized model for the AI use in computer crime investigations, taking into account explainability, auditability, and legal admissibility. The study employed the following methods: integrative stratification system of the typology of computer crimes, inter-normative comparative-procedural analysis of regulations, procedural modelling of the AI application scenario, synthesis of an optimized multi-agent model based on decomposition and normative analysis. The generalized results of the study showed that the integration of AI into forensics is limited by algorithmic opacity, regulatory fragmentation, and lack of procedural admissibility. The proposed optimized model with multi-agent architecture, explainability, auditability, and legal traceability minimizes algorithmic bias, formalizes forensic admissibility, and justifies the need to revise regulations and implement forensically explainable AI standards. The academic novelty is the formalization of an explainable AI-based forensic framework that integrates multi-agent analytics, legal traceability mechanisms, and standardized validation protocols, thereby advancing cognitive interpretability, regulatory admissibility, and institutional scalability of AI-evidence in criminal proceedings. The prospects for further research include conducting controlled empirical tests to verify the legal admissibility, interpretability, and procedural efficiency of the optimized procedural model.
Keywords:	Criminal Justice; Legal System; Rule of Law; Explainability; Legal Traceability
Source:	Journal of Theoretical and Applied Information Technology 30^th November 2025 -- Vol. 103. No. 22-- 2025
Full Text

Title:	A UNIFIED FRAMEWORK FOR DEVICE-AWARE WEB INTERFACE ADAPTATION BASED ON RESPONSIVE DESIGN AND HEURISTIC USABILITY EVALUATION
Author:	OLEH NI , OLEKSIY PISKAREV , YANA NI
Abstract:	In modern computer engineering, the efficiency of human-computer interaction depends on the adaptability of web interfaces across heterogeneous environments. This study analyzes Responsive and Adaptive Web Design approaches, evaluating their impact on usability, system performance, and interaction quality in multi-device ecosystems. Engineering considerations include rendering speed, resource efficiency, and cross-platform scalability. A scheduling interface prototype was implemented using HTML, CSS, and JavaScript, and optimized through media queries and front-end optimization techniques. Usability was assessed using the System Usability Scale (SUS) and heuristic evaluation. The results confirm that device-aware design models significantly enhance robustness and user satisfaction. This research highlights the importance of integrating UX best practices, adaptive layouts, and behavioral analytics. The proposed methodology synthesizes strategies from human-computer interaction, software architecture, and mathematical modeling. By uniting interdisciplinary insights, this work provides practical guidelines for developing scalable, user-centered web applications aligned with evolving digital system requirements.
Keywords:	Responsive Web Design (RWD), Adaptive Interfaces, Usability Adaptation, Heuristic Evaluation, System Usability Scale (SUS).
Source:	Journal of Theoretical and Applied Information Technology 30^th November 2025 -- Vol. 103. No. 22-- 2025
Full Text

Title:	COMBINING THE USER-CENTERED DESIGN METHOD WITH REQUIREMENT ENGINEERING TO IMPROVE USABILITY IN WEB-BASED APPLICATION
Author:	NAUVAL FAKSI ERLANSYAH, TANTY OKTAVIA
Abstract:	Usability is a crucial factor in ensuring the effectiveness and user satisfaction of digital systems, especially in high-stakes contexts such as e-Assessment. This study proposes an integrated approach that combines User-Centered Design (UCD) and Requirement Engineering (RE) to address UCD's limitations and improve the Usability of web-based applications, with a case study on an e-Assessment system. The research framework involves five independent variables—User Need, User Behavior, User Motivation, User Requirement, and Functional Requirement—hypothesized to influence Usability, which in turn affects the core usability dimensions of Learnability, Efficiency, Memorability, Error, and Satisfaction. The integrated method incorporates RE's structured processes—elicitation, analysis, specification, and validation—into UCD's iterative cycle. A prototype was developed and tested using quantitative methods, including path analysis and usability evaluation. The findings reveal that user behavior plays the most critical role in shaping Usability, while Usability itself positively impacts all core dimensions. This study contributes to the field by operationalizing the integration of UCD and RE into a comprehensive framework and validating its effectiveness in a web-based application. It demonstrates, through the e-Assessment case, how aligning structured requirements with user-centered practices can enhance the overall user experience.
Keywords:	User-Centered Design, Requirement Engineering, e-Assessment, Usability, System Usability Scale
Source:	Journal of Theoretical and Applied Information Technology 30^th November 2025 -- Vol. 103. No. 22-- 2025
Full Text

Title:	TRANSFORMERS IN ARABIC SHORT ANSWER GRADING: BRIDGING LINGUISTIC COMPLEXITY WITH DEEP LEARNING
Author:	WAEL HASSAN GOMAA, MENA HANY, EMAD NABIL , ABDELRAHMAN E. NAGIB, HALA ABDEL HAMEED
Abstract:	Automating the evaluation of Arabic short answers is a crucial step in advancing educational technology, as it enables rapid feedback, consistent scoring, and a significant reduction in educators’ workload. However, the structural richness and semantic complexity of Arabic—characterized by its extensive morphology, flexible word order, and diverse vocabulary—make reliable grading especially challenging. To address these difficulties, this study introduces a three-stage framework built upon fine-tuned transformer architectures. In the first stage, both the question and the learner’s response are encoded into dense semantic embeddings. The second stage applies comprehensive fine-tuning to a pre-trained transformer model, allowing it to capture task-specific nuances and better represent the intricate patterns of Arabic. In the final stage, a regression layer generates a numerical score, which is then compared against the human-assigned reference grade for evaluation. The proposed framework was rigorously tested on two benchmark datasets for Arabic short answer grading, AR-ASAG and Philosophy. Experimental results demonstrated strong performance, achieving Pearson correlation scores of 0.85 and 0.97, respectively, and outperforming previously reported state-of-the-art methods. These outcomes confirm the effectiveness of transformer-based models in handling the linguistic subtleties of Arabic while also demonstrating their scalability and adaptability across domains. Overall, the findings position fine-tuned transformers as a promising foundation for building accurate, efficient, and equitable automated grading systems in Arabic educational contexts.
Keywords:	Transformers, Arabic Short Answer Grading, Natural Language Processing, Deep Learning, Model Fine-Tuning
Source:	Journal of Theoretical and Applied Information Technology 30^th November 2025 -- Vol. 103. No. 22-- 2025
Full Text

Title:	THE IMPACT OF ARTIFICIAL INTELLIGENCE ALGORITHMS ON AUTOMATIC EDITING AND PROOFREADING OF TEXTS
Author:	OLEKSII SYTNYK, NATALIA KNYRIK, OKSANA TIUTIUNNYK, YEVGENYA BAZINYAN, TETIANA TSEPKALO
Abstract:	The relevance of the study is determined by the rapid development of artificial intelligence (AI) and the need to assess its impact on the quality of automatic editing of English texts in the global informational technologies (IT) environment. The aim was to identify and assess the impact of various AI algorithms on the efficiency and quality of automatic editing and proofreading of English texts in an international context. The objective of this study is to identify and analyze how various AI algorithms impact the efficiency and quality of automatic editing and proofreading of English texts in an international context. The study employed the following methods: experimental methods, content analysis, comparative analysis, and focus groups. The reliability was confirmed by using qualitative statistical analysis, Student’s t-test, chi-square test, and analysis of variance (ANOVA). The assessment was carried out in comparison with human editing. The texts included news, scientific articles, advertising texts, essays, and technical documentation. Readability analysis showed statistically significantly lower indicators of AI compared to editors (Flesch-Kincaid: 69.4 versus 75.2). The users noted the convenience and speed of intelligent technologies, but pointed to insufficient flexibility and stylistic problems. Analysis of variance confirmed the significant advantage of editors in quality (F=12.56, p<0.001, η²=0.45). The results obtained during the study showed a significant advantage of professional editors over modern AI algorithms in editing quality. At the same time, its use demonstrates the potential to accelerate the process, which is important for practical application. Prospects for further research include improving AI algorithms for stylistic editing and studying hybrid methods that combine automation with human expertise.
Keywords:	Natural Language Analysis; Neural Networks; Proofreading Quality; Text Processing; Text Proofreading
Source:	Journal of Theoretical and Applied Information Technology 30^th November 2025 -- Vol. 103. No. 22-- 2025
Full Text

Title:	LEVERAGING ARABERT WITH STATISTICAL WEIGHTING FOR SCALABLE UNSUPERVISED ARABIC TEXT SUMMARIZATION
Author:	WADEEA R. NJI, SURESHA, MOHAMMED A. S AL-MOHAMADI, AHMED R. A. SHAMSAN
Abstract:	The volume of Arabic digital text found in news platforms, educational resources, and social media continues to grow rapidly, making it increasingly difficult for users to extract key information in a timely manner. Automatic Text Summarization (ATS) provides an effective way to condense lengthy documents while preserving their essential meaning. However, progress in Arabic ATS remains limited because annotated datasets are scarce, Arabic-specific NLP resources are underdeveloped, and many recent language models demand high computational costs. Traditional summarization techniques also struggle to capture deeper sentence level semantics, which reduces the relevance and coherence of the generated summaries. To address these challenges, this study proposes a scalable unsupervised summarization framework that integrates TF-IDF weighting with AraBERT contextual embeddings to produce richer and more informative sentence representations. The model uses k-means clustering to identify thematic structure and selects representative sentences based on their similarity to cluster centroids. Maximal Marginal Relevance is then applied as a final step to reduce redundancy and maintain diversity across the selected content. Experimental results on the EASC dataset show that the weighted AraBERT representation achieves a ROUGE-1 score of 0.615, surpassing FastText, TF-IDF, and unweighted AraBERT. The findings demonstrate that integrating statistical term importance with contextual transformer embeddings provides an efficient strategy for enhancing summarization quality in low-resource Arabic settings. The proposed framework contributes a scalable, annotation-free alternative to supervised language models and provides new insight into representation weighting strategies for Arabic NLP.
Keywords:	Arabic Text Summarization; AraBERT, Weighted Embeddings; TF-IDF; Semantic Clustering, Unsupervised Learning.
Source:	Journal of Theoretical and Applied Information Technology 30^th November 2025 -- Vol. 103. No. 22-- 2025
Full Text

Title:	ENHANCING DATA PRIVACY AND OPTIMIZING LONG-DISTANCE DATA TRANSMISSION USING MACHINE LEARNING IN WIRELESS SENSOR NETWORK
Author:	SUBHRA PROSUN PAUL, SCHIN-SHIUH SHIEH, D. VETRITHANGAM, SIVA SHANKAR
Abstract:	Wireless Sensor Networks (WSNs) have emerged as a transformative technology that enables real-time data acquisition and communication across sectors such as healthcare, environmental monitoring, and smart transportation. WSNs have achieved notable progress in micro-electro-mechanical systems, low-power communication, and digital electronics, yet remain burdened by two ongoing issues: data privacy preservation and resource optimization for data transmission. The proposed research establishes a unified framework that combines Homomorphic Encryption (HE) with Federated Learning (FL) and Q-learning-based Reinforcement Learning (RL) to overcome existing challenges. The methodology is organized into three layers that support encrypted data acquisition, distributed model development, and optimized intelligent network routing. The Homomorphic Encryption system provides strong protection for encrypted sensor data, and the Federated Learning method prevents the transmission of raw information during model training. Through Q-Learning, the system achieves energy-efficient routing by dynamically adjusting paths based on network conditions. In addition to performance monitoring, the system provides real-time feedback loops that adaptively control encryption and routing parameters. The experimental tests conducted with NS-3 simulations and benchmark datasets showed that the HE-FL framework achieved a 0% success rate in data reconstruction under adversarial attacks, while maintaining acceptable model accuracy. Q-Learning-based routing improved network lifetime by 27% over LEACH, while delivering packets at 98.2%. The system operated with high computational efficiency using limited resources from microcontrollers. This research develops an extended, secure, and efficient WSN architecture that meets real-time requirements in healthcare and transportation networks. Edge intelligence benefits from this work because it integrates privacy-protecting algorithms and self-operating routing methods, overcoming standard WSN limitations. This framework demonstrates potential for general use in extensive security-critical sensor-based systems that need reliable information protection.
Keywords:	Homomorphic Encryption, Federated Learning, Q-Learning, WSNs, Privacy Preservation, Energy Efficiency
Source:	Journal of Theoretical and Applied Information Technology 30^th November 2025 -- Vol. 103. No. 22-- 2025
Full Text

Title:	INTEGRATION OF DATA FILTERING WITH HYBRID RSA DEEP LEARNING ALGORITHM FOR IOT DATA SECURITY AND CLASSIFICATION
Author:	ACHMAD FAUZI, SUCI RAMADANI, HUSNUL KHAIR, AKIM M H PARDEDE
Abstract:	The Internet of Things (IoT) presents significant challenges in maintaining data security, particularly in ensuring confidentiality while simultaneously detecting anomalies intelligently. Therefore, this study aims to develop a hybrid model that integrates the RSA algorithm for encryption and decryption with a Convolutional Neural Network (CNN) as the classification mechanism for IoT data. The research workflow includes RSA key generation and validation, encryption of IoT image data, decryption with the RSA private key, and classification using CNN with the log-sum-exp and softmax methods. Simulation results produced outputs of O₁ = −3945.78 with a probability of 0%, O₂ = −1972.89 with 0%, and O₃ = 0 with ≈100%, confirming that the input almost certainly falls into the file under attack class and demonstrating the model’s ability to preserve data confidentiality while achieving very high anomaly detection accuracy. The main contribution of this research is the development of a comprehensive approach to enhance the reliability of IoT systems against cybersecurity threats through the integration of RSA, which focuses on data confidentiality, and CNN, which ensures intelligent anomaly detection, so that the proposed model not only strengthens IoT security layers but also offers a practical solution that can be implemented to build more robust and adaptive security systems in the future.
Keywords:	CNN, Deep_Learning, Data_Security, IoT, RSA
Source:	Journal of Theoretical and Applied Information Technology 30^th November 2025 -- Vol. 103. No. 22-- 2025
Full Text

Title:	XMEDLLM: A FRAMEWORK FOR EXPLAINABLE MEDICAL INTELLIGENCE VIA LLMS IN EDGE-DRIVEN MCP SYSTEMS
Author:	K PRASUNA, T NAGAMANI, L MADHAVI DEVI, B KEERTHI SAMHITHA, MAHANTI SRIRAMULU, S SINDHURA
Abstract:	This Research proposes the design and test of an interpretable medical AI solution by incorporating Large Language Models (LLMs) like BioBERT and GPT-lite using edge-enabled Medical Cyber-Physical (MCP) systems. The model proposed is intended to resolve the most important problems of clinical AI such as interpretability, real-time inference, data privacy and scalability. The system allows the processing of medical data locally, on edge devices such as Jetson Nano and Raspberry Pi 4 via the use of LLMs, which not only keeps the latency to a minimum but makes sure that such privacy regulations as HIPAA are not broken. SHAP and LIME are used to give visual and textual understanding of AI decision algorithms, giving a great improvement to clinical trust. Clinician readiness to use the tool was observed in performance assessment performed in a simulated hospital environment which showed a high level of inference accuracy and excellent usability. The results of this research show that the AI models trained by LLM and deployed on the edges can be a scalable and safe solution to contemporary healthcare conditions.
Keywords:	Explainable AI, Large Language Models (LLMs), Edge Computing, Medical Cyber-Physical Systems, Clinical Decision Support
Source:	Journal of Theoretical and Applied Information Technology 30^th November 2025 -- Vol. 103. No. 22-- 2025
Full Text

Title:	QUANTUM-ENHANCED HYBRID XAI-LSTM FRAMEWORK FOR TELUGU RESTAURANT ANALYTICS AND RECOMMENDATION SYSTEMS
Author:	VENUGOPAL BOPPANA, SANDHYA PRASAD
Abstract:	This study examines the application of Explainable Artificial Intelligence (XAI) methodologies in a quantum-enhanced restaurant recommendation system designed for the evaluation of Telugu reviews. Our new framework uses the Natural Language Toolkit (NLTK) to process Telugu language, the Term Frequency-Inverse Document Frequency (TF-IDF) feature vectorization method, quantum-powered Ordering Points to Identify the Clustering Structure (OPTICS) algorithms, and the Long Short-Term Memory Recurrent Neural Network (LSTM-RNN) architecture with 100-cycle iterations. It also uses the whale optimization method to improve performance. Performance evaluation employs COSINE, DICE, and JACCARD similarity metrics across spatial, temporal, daily, and user feedback dimensions. The results show great results, such as 99.85% precision for Telugu preprocessing with 25,000 reviews and 99.76% precision for 28,456 unique tokens. Quantum-enhanced feature extraction reached 97.89% precision with 95.67% dataset coverage. Quantum LSTM networks reached 93.45% precision with 0.929 F1-scores. OPTICS clustering reached 91.23% precision with 0.934 AUC-ROC values, and whale optimization reached 92.34% precision with 0.945 AUC-ROC measurements. The integration of XAI led to amazing improvements, raising the overall system's accuracy from 85.5% to 95.6%, the users' confidence levels from 75.2% to 92.3%, and the ability to fix errors from 65.3% to 89.4%. Statistical analysis confirms the framework's exceptional efficacy in safeguarding Telugu cultural elements while offering clear recommendations. This illustrates the considerable potential of quantum computing when integrated with explainable AI to create culturally-sensitive, interpretable restaurant recommendation systems that propel multilingual recommendation research and lay the groundwork for future quantum-enhanced XAI studies.
Keywords:	Explainable Artificial Intelligence, Quantum Machine Learning, Restaurant Recommendation System, Telugu Natural Language Processing.
Source:	Journal of Theoretical and Applied Information Technology 30^th November 2025 -- Vol. 103. No. 22-- 2025
Full Text

Title:	RING THEORY-BASED AGRICULTURE RISK IDENTIFICATION USING HYBRID MODELS
Author:	M.SRIKANTH, M.CHANDRA NAIK, R.N.V.JAGAN MOHAN
Abstract:	Unrelenting plant diseases continue to pose a serious challenge to the agricultural sector around the world, undermining the production, food security, and economic stability of people. Conventional control systems are usually not timely assessed on their risks, and they do not consider the intricate element interdependence between environmental, biological, and crop-health components. This study tries to fill these shortcomings by developing a hybrid modelling system enhanced using a ring-theory-based influence mapping to enhance agricultural risk modelling. The research combines various data sets such as field surveys, environmental sensors that are based on the idea of IoT, remote sensing images, and past disease data. Statistical regression, decision trees, random forests, and artificial neural network models were trained as hybrid models simultaneously, optimized by ensemble methods and verified by cross-validation methods and interpretability methods such as SHAP and LIME. Findings indicate that the addition of the ring theory has a considerable positive impact on predictive performance: the accuracy, recall, and AUC-ROC increased by 4.2, 5 points, and 0.05 points, respectively, when compared to the performance of the models that were not augmented using the ring theory. In addition to the predictive gains, the framework provides early-warning signals, informs biosecurity measures, and assists with the development of resistant crops. The work presented in this study can add to the issue of agricultural resilience because it provides a mathematically informed and data-driven solution to the challenge of managing disease risks, which meets the ethical standards of AI and sustainable agriculture.
Keywords:	Agriculture, Risk Identification, Ring Theory, Hybrid Models, Disease Management, Agricultural Resilience
Source:	Journal of Theoretical and Applied Information Technology 30^th November 2025 -- Vol. 103. No. 22-- 2025
Full Text

Title:	SECURE SCADA FEDERATED INTELLIGENCE FRAMEWORK FOR CYBER SECURITY IN INDUSTRIAL NETWORK
Author:	YASIR A , KALAIVANI KATHIRVELU , MK. ARIF
Abstract:	The increasing frequency and sophistication of cyber threats in Supervisory Control and Data Acquisition (SCADA) systems pose significant challenges to industrial cybersecurity, particularly in decentralized and heterogeneous networks. Existing machine learning and optimization-based intrusion detection approaches suffer from inconsistent data representation, suboptimal feature selection, class imbalance, and poor generalization across federated nodes, leading to high false positives, missed attacks, and scalability limitations. To address these gaps, this study proposes Secure SCADA Federated Intelligence (SSFI), a unified framework integrating adaptive preprocessing, hybrid feature optimization, and federated attack classification. The Adaptive Feature Transformation (AFT) module improves data consistency and class balance, reducing preprocessing inconsistencies by 27.6% and enhancing minority class representation by 41.3%. The Adaptive Swarm-Bayesian Feature Optimization (ASB-FO) technique combines Particle Swarm Optimization (PSO) with Bayesian Optimization to select highly relevant features, achieving a 23.8% reduction in redundancy and a 16.2% improvement in classification accuracy. The Adaptive Federated Attack Classification (AFAC) module employs adaptive client-weighted aggregation to mitigate model divergence, resulting in a 32.5% reduction in global model instability and a 19.4% increase in detection accuracy. Experimental validation on benchmark SCADA datasets demonstrates that SSFI outperforms existing federated intrusion detection systems, achieving 96.8% accuracy, a 5.7% reduction in false positives, and a 14.9% improvement in precision. By addressing limitations in data preprocessing, feature optimization, and federated learning, SSFI introduces a novel, scalable, and privacy-preserving cybersecurity framework, representing a significant advancement in attack detection for industrial SCADA networks.
Keywords:	Cyber-Physical Systems, Federated Learning, Feature Selection, Optimization Techniques, Intrusion Detection, IOT-WSN attack, SCADA networks and CPS Security.
Source:	Journal of Theoretical and Applied Information Technology 30^th November 2025 -- Vol. 103. No. 22-- 2025
Full Text