		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
September 2025 | Vol. 103 No.18

Title:	PANTUNDUDA: EXPLAINABLE AI TO LANDMARK FACE RECOGNITION USING CANONICAL VARIATE ANALYSIS
Author:	AZREE NAZRI, OLALEKAN AGBOLADE, FAISAL AZIZ, ISZUANIE SYAFIDZA CHE ILIAS
Abstract:	The explainable face recognition (XFR) paradigm is concerned with the difficult challenge of deciphering the findings obtained by a facial matching system in order to shed light on the underlying rationale behind the selection of a specific identity in response to a given probe. The usefulness of biologically significant face cues in the context of gender classification is investigated in this study. The suggested statistical methodology involves detecting interpretable geometric features near anatomical markers, known as PANTUNDUDA. One classifier, consisting of 68 landmarks (68 -LM), was meticulously examined. An empirical investigation was carried out using a dataset of 100 facial images that were evenly divided among male and female subjects. The procedural framework seamlessly incorporates PCA, TPS, Procrustes superimposition, and Canonical Variate Analysis (CVA), with each component contributing independently to the extraction of comparable shape data from facial profiles. This method makes use of the designated landmark classifier. The findings highlight the importance of 20 facial landmarks as biologically interpretable markers for the classifier. These revelations have far-reaching consequences for XFR, particularly for landmark classifiers designed for gender-specific face comparisons. As a result, facial recognition algorithms have become more transparent and understandable.
Keywords:	PANTUNDUDA, AI, Variate Analysis, Facial Recognition
Source:	Journal of Theoretical and Applied Information Technology 30^th September 2025 -- Vol. 103. No. 18-- 2025
Full Text

Title:	APPLICATION OF KANSEI ENGINEERING METHOD FOR PRODUCT FORM DESIGN
Author:	FENGTAO ZHANG, ABDUL AZIZ BIN ZALAY， HAIXIA JIANG
Abstract:	With the development of social and the increasingly fierce market competition, users' perceptual needs have become a factor that can’t be ignored in product design, and people's perceptual cognition of products plays an increasingly important role in purchasing decisions. Kansei engineering（KE）as a design method that can quantify users' perceptual needs, transform them into specific design elements, and improve users' satisfaction with design, has attracted a lot of attention and research. this paper takes 2020 to 2024 as the research time span, summarizes the working principles, basic processes and methods of KE through systematic review of 221 publications and mining of 67 core literatures. This paper analyzes the current research status of KE from four aspects: the construction of user perceptual image lexicon, the deconstruction of product modeling, the construction method of KE model and the application of KE, and forecasts the future research. This research summarizes the research framework of Kansei Engineering in the field of product styling design, provides a data basis for future theoretical research on Kansei Engineering, and also provides research methodological assistance to more designers.
Keywords:	Kansei Engineering, Perceptual Image, Form Design, KE Model, Review
Source:	Journal of Theoretical and Applied Information Technology 30^th September 2025 -- Vol. 103. No. 18-- 2025
Full Text

Title:	THE IMPACT OF DIGITAL TECHNOLOGIES ON ACCOUNTING TRANSFORMATION
Author:	IRYNA SHCHYRBA, NATALIIA LOKHANOVA, NATALIІA HOLIACHUK, VALENTYNA SHEVCHENKO, YEVHENIIA KARPENKO
Abstract:	Digital technologies are significantly transforming accounting, changing the methods of collecting, processing, and analysing financial information on a global scale. The aim of the study is to assess the impact of digital technologies on accounting transformation in seven countries with different levels of digitalization. The research employed the following methods: quantitative approach, analysis of panel data for 2020-2024, and econometric modelling. The assessment was carried out using a fixed-effects model that takes into account institutional and technological factors of transformation. The results showed that the United Kingdom (UK) and Germany have the highest accounting transformation indicators: within 43.5-47.2 points and from 41.0 to 44.8 points, respectively. This is explained by a high level of digitalization (DigitalTechIndex over 80) and the full application of International Financial Reporting Standards (IFRS). At the same time, Ukraine, despite limited digital resources (Digital Technology Index within 36.2-49.7), demonstrates positive dynamics of transformation — from 28.0 to 33.2 points. This is determined by the implementation of IFRS and existing regulatory support. It was found that digital transformation significantly improves accounting practices in combination with the adoption of IFRS and the development of human capital. The results emphasize the need for strategic implementation of digital technologies in accounting, which has implications for politicians and financial institutions. The study can be useful for shaping digital development policies and modernizing financial reporting. Further studies should focus on industry analysis and studying the impact of artificial intelligence (AI) in accounting decisions.
Keywords:	Accounting, Digital Technologies, International Standards, Financial Reporting, Econometric Modelling, Transformation, Digitalization, Human Capital.
Source:	Journal of Theoretical and Applied Information Technology 30^th September 2025 -- Vol. 103. No. 18-- 2025
Full Text

Title:	MULTI-CORE VIRTUAL MACHINE PROCESSING IN DATA CENTRE WITH IMPROVED DATA SHARING AND SECURITY OF THE DATA ENVIRONMENT
Author:	J. MANIKANDAN, DR.UPPALAPATI SRILAKSHMI
Abstract:	The efficient placement of virtual machines (VMs) in cloud environments is crucial for optimizing resource utilization and achieving balanced workloads. This paper proposed a Multi-Core Virtual Data Centre Resource Sharing Security (MCVDC-RSS). The proposed MCVDC-RSS comprises data centers that consider the multi-core nature of physical machines and leverage vector heuristics to achieve balanced and efficient VM placement. MCVDC-RSS considers various vector heuristics, including L1-norm, L2-norm, cosine angle, and dot product, to evaluate the distance and similarity between VMs and physical machine resources. By incorporating these heuristics, the algorithm aims to minimize resource wastage and distribute VMs across physical machines in a balanced manner. To evaluate the effectiveness of MCVDC-RSS, we conducted simulations using both Amazon EC2 instances and the GoCJ dataset. The results demonstrate that MCVDC-RSS outperforms other algorithms in terms of resource wastage, overhead, active and overloaded pCPUs (physical CPUs), and overall system performance. Furthermore, MCVDC-RSS effectively manages compilation and execution plan times, CPU and memory utilization. The algorithm ensures efficient utilization of computational resources, leading to improved performance and workload management. The MCVDC-RSS provides a robust and efficient solution for VM placement in cloud environments. The proposed MCVDC-RSS model exhibits significant performance in the security of the data centers with the optimized utilization of resources, wastage of resources and increased efficiency
Keywords:	Data Centers, Multi-Core Environment, Virtual Machine, Security, Resource Sharing
Source:	Journal of Theoretical and Applied Information Technology 30^th September 2025 -- Vol. 103. No. 18-- 2025
Full Text

Title:	STREAM PROCESSING ALGORITHMS FOR UNSTRUCTURED DATA ANALYSIS
Author:	VIKTORIIA SHUTENKO, MYKOLA MEDVEDIEV, YURII DOROSHENKO, VALERIA DROMENKO, NATALIIA OMETSYNSKA
Abstract:	Relevance of the selected research The rapid growth of unstructured streaming data and the need for real-time decision-making necessitate the development of effective algorithms tailored for dynamic processing, semantic enrichment, and adaptive analytics. Purpose The purpose of the study is to develop an optimized mathematical model for the processing of unstructured data streams to support real-time decisions. Methods The study used the following methods: decomposition analysis, mathematical formalism, optimization modeling, reinforcement learning, simulation modeling.The research employed a diverse array of methodologies, including decomposition analysis, mathematical formalism, optimization modeling, reinforcement learning, and simulation modeling. Results obtained This study presents an optimized algorithm for real-time processing of unstructured data streams, derived from the decomposition of tested models and the formalization of a generalized mathematical framework. By integrating Kafka with Zero-Copy I/O, CEEMDAN-based preprocessing, ONNX-inference models, knowledge graph enrichment, and QoS-driven orchestration (Kubernetes, KEDA), the proposed solution achieved a 3.8× reduction in Latency (from 250.0 to 65.0 ms), increased Accuracy to 91.3%, and doubled Throughput (1100 to 2200 events/sec). Additionally, CPU Load was reduced by 27%, Memory Usage by 45%, while the Adaptability Score and Semantic Alignment Score improved from 0.52 to 0.88 and from 0.64 to 0.91, respectively. These results confirm the algorithm’s efficiency, scalability, and applicability in intelligent Decision Support Systems under real-time constraints. Scientific novelty of the study The scientific novelty of the research lies in the development of a mathematical model for streaming processing, characterized by dynamic ingesting, semantic enrichment, and optimization through reinforcement learning, all aimed at enhancing real-time productivity. Prospects for future research Further research involves validating the optimized algorithm within high-entropy streams, the scaling to multimodal datasets, and the enhancement through adaptive retraining and dynamic orchestration.
Keywords:	Kafka, Flink, Semantic Enrichment, Reinforcement Learning, RocksDB, CEEMDAN Decomposition, ONNX Inference.
Source:	Journal of Theoretical and Applied Information Technology 30^th September 2025 -- Vol. 103. No. 18-- 2025
Full Text

Title:	ENHANCING CLOUD DATA SECURITY WITH POLYNOMIAL ELLIPTIC CURVE ZERO KNOWLEDGE PROOF (POLYECC-ZKP)
Author:	E.JANSIRANI, DR.N.KOWSALYA
Abstract:	Cloud computing has revolutionized data storage and access, but it remains vulnerable to various security threats. Cryptographic approaches like Zero Knowledge Proof (ZKP) and Elliptic Curve Cryptography (ECC) have been widely used to address these issues. In order to improve cloud data security, this study presents a novel Polynomial Elliptic Curve Zero Knowledge Proof (PolyECC-ZKP) algorithm. By including polynomial functions into the ECC architecture, the suggested technique provides secure data authentication and strong encryption. We present a thorough analysis of the PolyECC-ZKP algorithm and evaluate its performance in comparison to other methods that have already been developed, such as Lattice-Based Zero Knowledge Proof (LZKP), Multi-Party Computation and Zero Knowledge Proof (MPC-ZKP), Hybrid Elliptic Curve Cryptography and Zero Knowledge Proof (HECCZKP), Hybrid Zero Knowledge Proof with ECC and ECDSA (Hybrid ZKP with ECDSA), and ECC. Scalability, quantum resistance, computation overhead, and security are the basis for the comparison. According to experimental findings, PolyECC-ZKP improves cloud security while requiring little computing power and is resistant to both conventional and quantum attacks. The results demonstrate PolyECC-ZKP's ability to emerge as a formidable contender for safe cloud settings.
Keywords:	Cloud Data Security, Polynomial Elliptic Curve Cryptography , Zero Knowledge Proof,Hybrid Cryptography Quantum-Resistant Algorithms, Multi-Party Computation, Lattice-Based Cryptography
Source:	Journal of Theoretical and Applied Information Technology 30^th September 2025 -- Vol. 103. No. 18-- 2025
Full Text

Title:	DETECTION OF BRAIN TUMOR AND PREDICTION OF SEVERITY ANALYSIS USING OPTIMIZATION DRIVEN 3D CONVOLUTION NEURAL NETWORK
Author:	Mr.BHASKAR MEKALA, Dr.NEELAMADHAB PADHY, Dr. P.KIRAN KUMAR REDDY
Abstract:	Accurate assessment of brain tumor severity plays a pivotal role in effective diagnosis and treatment planning. In this research, a comprehensive framework for brain tumor severity analysis is proposed, leveraging advanced preprocessing, segmentation, feature extraction and deep learning techniques. The preprocessing phase employs an adaptive bilateral filter to mitigate noise and enhance image quality. Subsequently, a segmentation approach utilizing slap swarm boosted rough fuzzy c-means optimally partitions the brain images, facilitating precise tumor localization. Feature extraction is performed using Mayfly optimization driven 3D Convolution Neural Network (MO-3DCNN), effectively capturing discriminative information from the segmented regions. The core innovation of this research lies in the development of a novel MO-3DCNN model, driven by the Mayfly optimization-derived features. This model enables a more comprehensive analysis by considering the spatial context of the brain tumor. The proposed MO-3DCNN model is trained to recognize intricate patterns within the segmented regions, enabling automated severity classification. Extensive experiments on a diverse dataset determine the superiority of the proposed methodology over existing approaches. Moreover, the proposed MO-3DCNN model outperforms in accurately stratifying brain tumor severity, highlighting its potential as a valuable clinical tool.
Keywords:	Convolution Neural Network, Rough Fuzzy C-Means, Slaps Swarm Optimization, Mayfly Optimization, And Adaptive Bilateral Filter.
Source:	Journal of Theoretical and Applied Information Technology 30^th September 2025 -- Vol. 103. No. 18-- 2025
Full Text

Title:	IMPROVING USER EXPERIENCE WITH DIGITAL TRANSFORMATION PRACTICES IN ACADEMIA
Author:	NAJWA H SAMRGANDI
Abstract:	Digital transformation in higher education fundamentally reshapes how universities function and interact with students, faculty, and external stakeholders. Beyond technological upgrades, this transformation has direct implications for user experience (UX), influencing satisfaction, engagement, and retention. This study conducts a comparative analysis of digital transformation practices and their impact on UX at Carnegie Mellon University’s Digital Transformation and Innovation Center (CMU DTIC) and the University of California, Irvine’s Center for Digital Transformation (UCI CDT). A comparative case study approach was employed, examining website content, publications, and digital initiatives from both centers. Key dimensions included research focus, projects, faculty structure, collaborations, funding sources, and the integration of UX principles. The findings indicate that CMU DTIC adopts a predominantly technology-driven and solution-oriented approach, emphasizing responsible artificial intelligence (AI), machine learning (ML), and rapid-response initiatives. While this strategy produces technically advanced outcomes, it sometimes limits the depth of user-centered design integration. By contrast, UCI CDT follows a strategic and long-term perspective, prioritizing strategic digital transformation, leadership, and societal impact, though its initiatives occasionally lack strong practical UX application. Both institutions demonstrate strong collaborations and diverse funding models, with CMU leveraging a combination of industry collaborations and external funding sources, while UCI emphasizes internal stability. Overall, the analysis demonstrates that successful academic digital transformation is less dependent on the level of technological sophistication than on the adoption of inclusive, user-centered practices that enhance engagement, accessibility, and adoption. A balanced framework that integrates CMU’s technical responsiveness with UCI’s strategic, interdisciplinary orientation can optimize both research outcomes and practical impact within higher education.
Keywords:	Academia, Comparative Study, Digital Transformation, Interdisciplinary Collaboration, User Experience.
Source:	Journal of Theoretical and Applied Information Technology 30^th September 2025 -- Vol. 103. No. 18-- 2025
Full Text

Title:	THE SECURE DYNAMIC WIRELESS SENSOR NETWORK
Author:	ASO AHMED MAJEED, YALMAZ NAJM ALDEEN TAHER, HOGER K. OMAR, KAWA M KAKY
Abstract:	Wireless Sensor Networks (WSNs) vary in size by application and are often deployed in uncontrolled areas, making them vulnerable to attacks, especially on routing protocols. Due to weak transmission security, messages can be easily intercepted or altered. Thus, efficient key management is essential to reduce these risks. There are many challenges to securing key management, such as key distribution, routing algorithms, overhead, scalability, efficiency, and time consumption for encryption and decryption. Thus, it is challenging to create efficient security protocols while decreasing costs. The proposed scheme is more adaptive and secure because it works like a one-way function to save capacity and time execution compared with other schemes, and it achieves the security goals (integrity, authentication, confidentiality, data refresh, and scalability). Finally, it prevents the message from being repeated compared with another scheme, avoiding the adversary guessing the keys and penetrating the network.
Keywords:	Cryptography, Key Management, Security, Wireless Sensor Network, Data Sequence.
Source:	Journal of Theoretical and Applied Information Technology 30^th September 2025 -- Vol. 103. No. 18-- 2025
Full Text

Title:	FRAMEWORK FOR DATA INTEGRATION IN CROSS-DOMAIN PERVASIVE ENVIRONMENTS: LIGHTWEIGHT AND SECURE RESTFUL API
Author:	AJAI GOPAL BHARTARIYA, S. K. SINGH, AJAY KUMAR BHARTI
Abstract:	The digital revolution has compelled organizations to operate within increasingly complex digital environments. In such contexts, effective coordination is essential to ensure seamless interaction among diverse systems and stakeholders. This paper presents a structured framework for API adoption in business settings, developed through an extensive review of existing literature on API adoption strategies. Emphasizing the growing significance of RESTful APIs in today’s interconnected world, the study illustrates how APIs serve as a critical enabler for integrating heterogeneous technologies, systems, and platforms. By leveraging APIs, organizations can unlock new business opportunities, optimize operational efficiency, and deliver innovative services to customers. The central theme of this research is the development of a framework for data integration using RESTful APIs, with a particular focus on security. The paper highlights the pivotal role of security in API-based integration, examining authentication and authorization mechanisms alongside best practices for safeguarding data privacy and integrity. In addition, it explores monitoring and logging practices, as well as developer testing approaches, underscoring the importance of proactive issue detection and resolution to ensure the reliability and performance of APIs. The framework is flexible enough to enable organizations to focus on the specific areas that need their attention because digital organization structures differ in their degree of maturity. In order to address security, usability, real-time capabilities, and monitoring, this research study offers a thorough overview Framework for Data Integration in Cross Domain Pervasive Environments using Lightweight and Secure RESTful API.
Keywords:	RESTful, API, Security, Integration, Lightweight, Domain.
Source:	Journal of Theoretical and Applied Information Technology 30^th September 2025 -- Vol. 103. No. 18-- 2025
Full Text

Title:	A UNIFIED APPROACH COMBINING FUZZY LOGIC AND KERNEL-BASED TWIN PROXIMAL SVM FOR DETECTING LUNG CANCER
Author:	VIRINCHI PUVVADA , PRABAKERAN SARAVANAN
Abstract:	Lung cancer is the leading cause of cancer-related fatalities globally, emphasizing the importance of early detection to improve survival rates. Traditional diagnostic methods often struggle with issues of sensitivity and specificity, which calls for the development of more advanced techniques. This study introduces an improved method for detecting lung cancer using CT images, combining deep learning, fuzzy logic, and Twin Proximal Support Vector Machine (TPSVM). The proposed hybrid model applies fuzzy logic for feature fuzzification, uses the VGG19 model for transfer learning to extract features, and implements TPSVM for classification, further optimized with the Dolphin Echolocation Optimizer (DEO). When tested against conventional classifiers such as decision trees, Naive Bayes, and k-nearest neighbors (KNN), the model showed exceptional performance across various evaluation metrics. This approach offers a blend of high accuracy and computational efficiency, providing an effective tool for precise and early lung cancer diagnosis, while also contributing to the advancement of AI-based medical imaging technologies.
Keywords:	Deep Learning, Artificial Intelligence, Medical Image Diagnosis, Fuzzy Logic, Computer Vision.
Source:	Journal of Theoretical and Applied Information Technology 30^th September 2025 -- Vol. 103. No. 18-- 2025
Full Text

Title:	EFFICIENT MULTI-CLASS ATTACKS CLASSIFICATION FOR INTRUSION DETECTION USING WRAPPER-BASED FEATURE SELECTION
Author:	KURRA UPENDRA CHOWDARY, BHAVNA BAJPAI, V ANJANI KRANTHI, B MADHAV RAO, B VEERAMALLU, G SATYANARAYANA
Abstract:	Cloud computing's massive uptake across industries has resulted in huge advancements, but it has also introduced massive security holes. Intrusion detection systems are crucial for identifying and reducing risks in cloud environments since these weaknesses make them accessible to a wide range of harmful actions. Nevertheless, classifier performance could be negatively impacted by the abundance of irrelevant and superfluous features included in the massive amounts of modern network traffic data stored in the cloud. moreover, processing complexity rises with such massive information, thus reducing the efficacy of IDS. This study presents a hybrid feature selection method that integrates two bio-inspired algorithms: Particle Swarm Optimization (PSO) and Genetic Algorithm (GA) to address these challenges. This hybrid approach optimizes function selection, ensuring a more efficient search for relevant features, which can subsequently be utilized to train a Random Forest (RF) classifier. Furthermore, to address the imbalance in elegance, the observation incorporates a hybrid resampling procedure associated with the Synthetic Minority Over-sampling Technique (SMOTE) for the minority class. This technique improves the true positive rate (TPR) while simultaneously reducing the false positive rate (FPR), hence enhancing IDS performance.. The proposed approach become evaluated the usage of the CICDDoS2019 dataset, displaying superior overall performance in multiclass category. The hybrid IDS outperformed several different classifiers, inclusive of Stacking, Extra trees, Multi-Layer Perceptron, XGBoost, K-Nearest Neighbors, Logistic Regression, Naïve Bayes, Support Vector Machine, and Decision Tree.The method also showed consistent performance across more than one evaluation metric, which was better than the best methods at the time.
Keywords:	Cloud Computing, Intrusion Detection System, Particle Swarm Optimization, Genetic Algorithm (GA), Synthetic Minority Oversampling Technique , Class Imbalance, Cicddos2019 Dataset
Source:	Journal of Theoretical and Applied Information Technology 30^th September 2025 -- Vol. 103. No. 18-- 2025
Full Text

Title:	SNOMEDTM: A NOVEL TRANSFORMER-BASED ARCHITECTURE FOR ADVERSE DRUG EVENT EXTRACTION FROM CLINICAL TEXT
Author:	SALISU MODI, KHAIRUL AZHAR KASMIRAN, NURFADHLINA MOHD SHAREF, MOHD YUNUS SHARUM
Abstract:	Extraction of adverse drug event (ADE) mentions and their attributes and relations within electronic health records are crucial for adequate pharmacovigilance studies and drug safety surveillance. Transformer-based large language models (LLMs) have recently shown promising results in this research. However, clinical LLMs are few and have a limited number of parameters. General LLMs are domain-agnostic models developed for varying NLP tasks. However, due to the domain-specific nature of ADE extraction with ambiguous, polysemous and infrequent entities, general LLMs lacking prior medical knowledge have been observed to perform sub-optimally in handling these complex situations within clinical narrative documents. Consequently, researchers further pre-train the general models on domain-related knowledge before finetuning them for the downstream clinical tasks. Nevertheless, this approach is associated with several risks. The model may overfit when the domain-specific data is small. In addition, catastrophic forgetting may occur. To propose a new architecture tailored to extract ADEs called the SNOMED Transformer Model (SNOMEDTM) pre-trained from globally standard medical knowledge bases. The process is in two phases: A new transformer architecture was designed and pre-trained on the medical terminology-based SNOMEDCT and MedDRA. The model is tuned using fine-tuning and soft prompt tuning for multi-task ADE concept and relation extraction tasks. This study experimented with two tuning strategies, frozen and unfrozen model parameters. The model’s performance was evaluated using the TAC 2017 and n2c2 2018 clinical challenge datasets. On TAC 2017, the proposed model outperformed the five compared transformer-based models and the top five systems contributing to the TAC 2017 challenge for fine-tuning. On n2c2 2018, the model outperformed GatorTron-base for soft prompting with unfrozen model and JNRF systems. This research demonstrates the potential of incorporating prior medical knowledge into LLMs tailored for clinical research.
Keywords:	Adverse Drug Event, Fine-tuning, Soft Prompt Tuning, Prior Knowledge, Pretraining.
Source:	Journal of Theoretical and Applied Information Technology 30^th September 2025 -- Vol. 103. No. 18-- 2025
Full Text

Title:	FUSION OF GLOBAL OPTIMIZED CONVOLUTIONAL NEURAL NETWORK WITH DEEP FLEXIBLE NETWORK FOR DIGIT RECOGNITION
Author:	PRIYANGA K.K, S. SABEEN
Abstract:	Digit recognition refers to the process of identifying and classifying digits, especially from handwritten or printed sources. It involves analyzing the features of each digit to match it to the correct number. Combining a Global Optimised Convolutional Neural Network (GOCNN) with Deep Flexible Network (DFEN) has resulted in a unique and powerful digital identification system. It can manage scalability, accuracy, and flexibility. This hybrid model outperforms the others by combining the high dynamic flexibility of DFEN for handling complex patterns and shifting the structures with the robust feature extraction capabilities of CNNs. Global Optimized Convolutional Neural Networks provide consistent feature extraction across several datasets. The Deep Flexible Network acts as a counterweight by continuously learning and adapting to new input, including distorted handwriting. The convolutional layers of the proposed systems employ two complex optimization approaches, including gradient-based corrections and adaptive learning rates. This strategy enhances the model's dependability by enabling the extraction of consistently high-quality features. This fusion strategy allows the model to interpret both coarse-grained and abstract data by combining the information at several levels. As it is more precise and efficient in digit recognition, its design is appropriate for real-time use. The combination of technologies, including digital forensics, computerised handwriting analysis, secure authentication methods, and word recognition in scanned documents, benefits several disciplines. Even though it is resilient and flexible enough to handle complex data and adapt to various input patterns, strong identification performance is maintained. This study provides a foundation for future research on hybrid architecture, focusing on the balance between accuracy and flexibility in pattern recognition.
Keywords:	Accuracy, Convolutional Neural Network (CNN),Deep Flexible Networks (DFEN), Gradient, Secure authentication.
Source:	Journal of Theoretical and Applied Information Technology 30^th September 2025 -- Vol. 103. No. 18-- 2025
Full Text

Title:	DEVELOPING AN ENSEMBLE OF NEURAL NETWORKS TO ASSESS OPERATIONAL RISKS TRIGGERED BY EMPLOYEE ERRORS OR MISCONDUCT
Author:	EKATERINA CHUMAKOVA , DMITRY KORNEEV , MIKHAIL GASPARIAN , ANDREY PONOMAREV , ILIA MAKHOV
Abstract:	The article addresses operational risk management in credit institutions, specifically risks arising from employee actions. It examines personnel impact on the continuity of business processes as one of the four sources of operational risk events. For example, insufficient employee qualifications can lead to various losses for a credit organization. Consequently, one potential approach to minimizing operational risks related to personnel activities is the use of artificial intelligence technologies in developing tools for assessing the criticality level of operational risks triggered by employee errors or misconduct. The study aims at developing an intelligent system for preventive monitoring of critical conditions in business processes caused by staff actions or inaction to prevent operational risks. To achieve this objective, the study analyzes professional and personal criteria for employee evaluation, their impact on business processes, and accumulated statistical indicators. A general structure for a business process status indication system is proposed, organized according to a modular principle. Feedforward neural networks (NNs) are suggested as modular system components. The article outlines the main data streams feeding into the inputs of NNs and compares different NN models used within each system module. The results can be used by credit institutions to enhance operational efficiency by adopting new approaches to assessing the criticality of operational risks associated with employee behavior.
Keywords:	Operational Risk, Personnel Actions, Artificial Neural Network, Machine Learning, Feedforward Neural Network.
Source:	Journal of Theoretical and Applied Information Technology 30^th September 2025 -- Vol. 103. No. 18-- 2025
Full Text

Title:	UTILIZING FUZZY-IDENTITY-BASED ENCRYPTION WITH PROXY-RE-ENCRYPTION FOR DATA SHARING
Author:	JIBIN JOY , DR. S. DEVARAJU , DR. J. RAMKUMAR
Abstract:	Replication of information also means that efficiency of storing cost is assured since only a copy of information is kept according to the different types of variations. This has been necessitated even further by the fact that, the world information has increased at a greater pace. The sharing of data in the cloud is assessed using Fuzzy-Identity Based-Encryption with Proxy-Re-Encryption (FuzzyIBE PRE) in the new research. The team of researchers is interested in identifying the answers to the question of discovering the effectual mechanism of sharing applied by users of cloud by embracing the benefits of PRE FuzzyIBE. FuzzyIBE PRE proves that it can use open access control scheme by maintaining the privacy of the information. In order to realize this research paper, the research shall provide a critical narration of all the available literatures on the attribute-based encryption and the proxy re encryption in specific with regard to the fuzzy identity description. This will be followed by the running of the FuzzyIBE PRE on real life assessments to determine the degree of effectiveness FuzzyIBE is going to have when it comes to actual implementation of the application cloud sharing data. The study objective will be to get an understanding of the secure privacy based data exchange functions in clouds.
Keywords:	Fuzzy-Identity-Based-Encryption with Proxy-Re-Encryption (FuzzyIBE-PRE), Fuzzy-Identity-Based-Encryption (FIBE), Media-Access-Control (MAC), Information-Management-Table (IMT), Non-Volatile memory (NVM)
Source:	Journal of Theoretical and Applied Information Technology 30^th September 2025 -- Vol. 103. No. 18-- 2025
Full Text

Title:	AN OPTIMIZED ENSEMBLE APPROACH FOR ALZHEIMER’S DISEASE DETECTION: INTEGRATING GRADIENT BOOSTING TECHNIQUES WITH FEATURE SELECTION AND HYPERPARAMETER TUNING
Author:	T MAHA LAKSHMI, G SUDHAVANI, K RAMANJANEYULU, G V PRASANNA ANJANEYULU, LAVANYA GOTTEMUKKALA, RUHISULTHANA SHAIK
Abstract:	Early diagnosis and correct diagnosis of Alzheimer's disease (AD) are crucial for intervention on time. The present research proposes E3-Boost, a new ensemble learning approach that is effective in improving the early diagnosis of Alzheimer's disease (AD). The novelty of the research lies in the unified integration of Gradient Boosting Machine (GBM), Extreme Gradient Boosting (XGBoost), and Light Gradient Boosting Machine (LightGBM), which together ride on the strength of combinations of boosting algorithms to enhance predictive performance. One of the main novelties is L1L2-FS, a hybrid L1 (Lasso) and L2 (Ridge) regularization feature selection approach that, with respect to the conventional practice, enables even more efficient identification of features as well as their redundancy removal. Not only is the model increased in interpretability, but so is its capability for generalization. Additionally, the methodology leverages SMOTE for handling imbalanced classes as well as for outlier detection on the basis of Z-Scores to guarantee excellent data quality. Another milestone is the implementation of Optuna-based hyperparameter tuning through Bayesian Optimization and Tree-structured Parzen Estimators (TPE), which iteratively optimizes model performance to achieve a startling accuracy rate of 98.82%. The last model realizes a substantial boost in accuracy (97.85%), recall (96.92%), F1-score (96.88%), and lowered RMSE (0.1023), outperforming conventional ML and deep learning models. In contrast to computationally intensive deep learning methods, E3-Boost presents a computationally light solution with high predictive power, which is well suited for real-world clinical applications. The most important innovation is three pillars they are L1L2-FS, a hybrid feature selector that merges Lasso's sparsity and Ridge's resilience to collinearity; E3-Boost, the first ensemble that brings together GBM, XGBoost, and LightGBM through dynamically weighted meta-learning; and third one is Optuna-TPE optimization, automating hyperparameter tuning with Bayesian-adaptive sampling. This triad fills crucial gaps in interpretability, scalability, and reproducibility for clinical AI.
Keywords:	E3-Boost, Alzheimer’s Disease Prediction, Ensemble Learning, Feature Selection, SMOTE, Hyperparameter Tuning
Source:	Journal of Theoretical and Applied Information Technology 30^th September 2025 -- Vol. 103. No. 18-- 2025
Full Text

Title:	EXTENDING THE THEORY OF PLANNED BEHAVIOR: THE ROLE OF INTRINSIC AND EXTRINSIC MOTIVATIONS IN PREDICTING DIGITAL GAMING BEHAVIOR AMONG COLLEGE STUDENTS
Author:	WANG XIAODAN, AINI AZEQA MA’ROF, HASLINDA ABDULLAH, ZEINAB ZAREMOHZZABIEH, YAN TANG
Abstract:	The purpose of this study was to enhance the Theory of Planned Behavior (TPB) by integrating constructs from Self-Determination Theory (SDT), specifically intrinsic and extrinsic motivation, to better predict digital gaming intentions and behaviors among college students. A cross-sectional design was employed, and data were collected from a sample of 850 Chinese college students using a multi-stage cluster random sampling method. Structural equation modeling was used to test the extended model and assess the predictive relationships between motivational and cognitive factors influencing digital gaming behavior. The findings indicated that attitudes toward gaming, subjective norms, and perceived behavioral control significantly predicted the intention to engage in gaming, which, in turn, predicted actual behavior. Intrinsic motivation had a stronger impact on attitudes and perceived behavioral control than extrinsic motivation, and it influenced behavior indirectly through serial mediation pathways. The results also showed gender did not significantly affect gaming behavior. The findings suggest that game designers and educators should prioritize intrinsic motivational elements—such as autonomy, challenge, and relatedness—when developing game-based learning tools. Tailored interventions based on students’ academic levels and geographic backgrounds are also recommended to enhance engagement. This study extends the TPB by integrating motivational dimensions from SDT, offering a more comprehensive framework for understanding technology-related behaviors. It also demonstrates the value of multilevel serial mediation in explaining digital engagement. By addressing both psychological and contextual factors, the extended model provides new insights into how youth adopt and engage with digital gaming technologies.
Keywords:	Theory of Planned Behavior, Intrinsic Motivation, Extrinsic Motivation, Digital Gaming Behavior, Self-Determination Theory
Source:	Journal of Theoretical and Applied Information Technology 30^th September 2025 -- Vol. 103. No. 18-- 2025
Full Text

Title:	ADAPTATION OF LOSS RECOVERY MECHANISMS FOR IMPROVING SCALABILITY AND QUALITY OF SERVICES IN IOT NETWORKS
Author:	ABOURRICHE SAMIRA , ZYANE ABDELLAH , GHAMMAZ ABDELILLAH
Abstract:	Over the past decade, the Internet has undergone significant transformations in various sectors such as healthcare, transportation, and the environment. These developments are built upon the TCP/IP model protocols, a four-layer architecture where the transport layer plays a crucial role in enabling Internet communication and managing Quality of Service (QoS). This evolution spans from the original ARPANET design to the emergence of Web 3.0, including Web 1.0 and Web 2.0. It has enabled the development of new communication paradigms such as the Internet of Things (IoT), which seeks to connect physical objects and devices across diverse domains within a unified infrastructure, supporting real-time monitoring and control [1].As a result, IoT has become a leading technology in multiple sectors, with connected devices projected to increase from 30.7 trillion in 2020 to 75.4 trillion by 2025 [2]. This massive number of connected things introduces numerous challenges, particularly in terms of QoS and scalability, which present obstacles for IoT and, more specifically, Machine-to-Machine (M2M) networks.The main objective of this paper is to provide a comprehensive contribution toward implementing mechanisms for autonomous scalability management and Quality of Service (QoS). The approach integrates Transport Layer loss recovery protocols from the TCP/IP model within the middleware layer of IoT networks.. The outcome of this work proposes a new architecture for the existing IoTScal approach, incorporating additional components to simulate loss recovery mechanisms. This enhancement boosts the success rate of e-Health traffic from 95% to 99.99%, delivering a clear and significant improvement over the existing reference approach.
Keywords:	Quality Of Service,Internet Of Things,Scalability,M2M Networks,Loss Recovery Mechanisms
Source:	Journal of Theoretical and Applied Information Technology 30^th September 2025 -- Vol. 103. No. 18-- 2025
Full Text

Title:	OPTIMIZING SUSTAINABILITY IN SMART MICRO GRIDS: NATURE-INSPIRED MULTI-OBJECTIVE ECONOMIC EMISSION DISPATCH MODELS
Author:	V. SAI GEETHA LAKSHMI, M. DEVIKA RANI , R BALUSULARAO MALUKURT, RAMAVATH SRINIVAS, SRIKANTH KILARU
Abstract:	In order to prevent transmission losses and keep the electricity on continuously, the micro grid systems generate and distribute power on a smaller scale to a limited geographic area. With the goal of reducing air pollution caused by burning fossil fuels, it has been standard practice to use renewable energy sources (RES). Distribution energy resource (DER) sizing with fuel cost minimization is to focus on economic load dispatch (ELD). Optimal sizing of DERs sources is accomplished via emission dispatch, which minimizes atmospheric pollutants. For the best results in reducing fuel costs and pollution, a multi-objective Combined Economic-Emission Dispatch (CEED) is used to size distributed energy resources (DERs). Using a newly-developed unique technique enhanced golden jackal optimization algorithm (EGJO) this paper executes all three phases that is ELD (economic load dispatch), ED (emission dispatch), and CEED. In this study, four different situations using DER load sharing are examined on hourly basis. The suggested method's efficacy is confirmed by comparing the findings to other newly created bio-inspired algorithms. Additional statistical analysis, including a two-way ANOVA test, is carried out to demonstrate that the suggested method outperforms than the other optimization strategies.
Keywords:	Bio-Inspired Algorithms, Micro Grid, Multi- Objective Combined Economic-Emission Dispatch, Renewable Energy Resources.
Source:	Journal of Theoretical and Applied Information Technology 30^th September 2025 -- Vol. 103. No. 18-- 2025
Full Text

Title:	QUANTITATIVE BENCHMARKING AND CROSS-MODAL ANALYSIS OF DEEP LEARNING, MACHINE LEARNING, AND BIOSENSOR FRAMEWORKS FOR EARLY COLORECTAL CANCER DIAGNOSIS AND PROGNOSIS
Author:	DIVYA MIDHUNCHAKKARAVARTHY, G MUNI NAGAMANI, V LAKSHMAN NARAYANA
Abstract:	This text is an iterative major study that discusses colorectal cancer (CRC), which is also a formidable malignancy for most countries in the world, with early detection making it a boon for improving survival rates. This is an empirical review of recent works that apply machine learning (ML), deep learning (DL), and multi-omics-based biosensor systems toward CRC diagnosis and prognosis. In the field of Information Technology (IT), this research fills a fundamental requirement for the development of deployable healthtech systems by providing a standardized, computationally grounded framework for performance benchmarking. This framework allows for fair comparisons and reproducibility. This review, unlike other previous reviews, applies a six-metric evaluation framework accuracy, precision, recall, RMSE, inference delay, and computational complexity to benchmark models systematically between imaging and non-imaging modalities. Hybrid models such as CMNV2, DeepCPD, and MACGAN achieved classification accuracies exceeding 99%, with CMNV2 proving most effective at 99.95% and perfect recall for histopathological datasets. Furthermore, designs utilizing a transformer architecture like MLPFormer and MSNet outperformed baseline models in segmentation tasks, improving Dice scores of 3-5%. Among these mutants, however, genomic and survival models for example DeepSEA Further enhance this prediction with good interpretability but have moderates performance (CIindex ~0.71). The visual analytics using the above medium like violin plots, heat maps, and correlation will reveal the performance trends and the expression of trade-offs made between accuracy and model complexity. The paper, therefore, establishes a high-resolution benchmarking map that informs one on model selection depending on application needs ranging from polyp detection to survival predictions. Future research directions are identified toward the goal of having explainable and lightweight multi-modal architectures and validation in multi-center prospective clinical trials in process.
Keywords:	Colorectal Cancer, Deep Learning, Machine Learning, Diagnostic Modeling, Survival Prediction, Scenarios.
Source:	Journal of Theoretical and Applied Information Technology 30^th September 2025 -- Vol. 103. No. 18-- 2025
Full Text

Title:	INTELLIGENT NETWORK INTRUSION DETECTION USING ENSEMBLE CLASSIFIER AND RELIEFF WITH PROBABILISTIC OPTIMIZATION OF SIMULATED ANNEALING
Author:	SVSV PRASAD SANABOINA, Dr.M CHANDRA NAIK, Dr.K RAJIV
Abstract:	Recently, the Intrusion Detection System (IDS) analyzes Internet of Things (IoT) network packets and further reports security violations to the authorized system administrator. The reports become unmanageable in the larger networks, and the existing systems suffer from a high false-alarms-rate. In order to address this concern, an efficient intelligent automated IDS is proposed based on the ReliefF with Probabilistic Optimization of Simulated Annealing (RPOSA) algorithm and ensemble classification model. In the initial phase, the label encoder and Z-score technique are applied on the NSL-KDD and UNSW-NB15 databases to normalize feature categories and eliminate outliers. Furthermore, the optimal intrusion attributes are selected from the pre-processed data using RPOSA algorithm. In this algorithm, the ReliefF method is initially employed for ranking the attributes based on its relevance to the target variable, and further, the simulated annealing optimization algorithm is applied for searching optimal attributes utilizing the ReliefF rankings. These selected optimal intrusion attributes are fed into the Stacked Ensemble Classification (SEC) model for efficient and accurate classification of intrusion attacks, and it comprises three models: Deep Neural Network (DNN), logistic regression, and Gradient Boost Decision Tree (GBDT). The RPOSA-SEC model obtained an impressive classification accuracy of 99.37% and 99.52% on the databases of NSL-KDD and UNSW-NB15, particularly in the case of both multi and binary classification.
Keywords:	Ensemble Classification Model, Internet of Things, Intrusion Classification, Label Encoder, ReliefF, Simulated Annealing, Z-score Technique
Source:	Journal of Theoretical and Applied Information Technology 30^th September 2025 -- Vol. 103. No. 18-- 2025
Full Text

Title:	ENSEMBLE MACHINE LEARNING AND SWARM INTELLIGENCE FOR ROBUST HEART DISEASE PREDICTION: LEVERAGING THE POWER OF CUCKOO HARMONY OPTIMIZATION
Author:	G. V. S. N. R. V. PRASAD, CH. DEVI CHAITANYA
Abstract:	Heart disease is still a major worldwide health issue that demands the creation of precise and effective predictive models. This study seeks to provide an innovative approach for predicting cardiac disease through ensemble machine learning techniques using swarm intelligence. In this research the risk of heart disease was split into four separate groups using K-Means clustering applied on the Cleveland Cardiovascular Disease dataset from the University of California, Irvine: no risk, low risk, medium risk, and high risk. We examine several machine learning techniques including Logistic Regression, Naive Bayes, Decision Tree, Multilayer Perceptron, Support Vector Machine, K-Nearest Neighbors, Random Forest, CNN—both with and without the Cuckoo Harmony approach concurrently. The CNN model attaining almost ideal performance (97% accuracy, 100% precision, and 100% ROC-AUC) shows a notable increase in prediction accuracy and resilience by means of swarm intelligence. Our results show that this development is really significant. The results of this work show that it is successful to include advanced machine learning and statistical inference into the process of risk of heart disease prediction and opens the path for further developments in the discipline of healthcare data analysis.
Keywords:	Heart Disease Prediction, Machine Learning, Swarm Intelligence, Cuckoo Harmony Algorithm, Ensemble Learning.
Source:	Journal of Theoretical and Applied Information Technology 30^th September 2025 -- Vol. 103. No. 18-- 2025
Full Text

Title:	ADVANCING BIOMETRIC IMAGE GENERATION AND VERIFICATION WITH DEEP LEARNING-BASED GENERATIVE AI METHODS
Author:	SHANTHI PANNALA, Dr B SATEESH KUMAR
Abstract:	This study introduces a novel biometric authentication framework integrating GenBio-Net, BioSynth-VerifyNet, SecureGen-ID, and Ethical-BioGuard into a unified system capable of achieving an accuracy of up to 98.5% on synthetic and real fingerprint datasets. The proposed approach addresses critical challenges in generative biometric security, privacy-preserving verification, and ethical usage policies. Our framework synthesizes biometric data using advanced generative modeling, enhances verification robustness with multimodal learning, and incorporates federated learning for secure deployment. Experimental results on benchmark and synthetically generated fingerprint datasets demonstrate superior performance over recent state-of-the-art methods in terms of accuracy, false acceptance rate (FAR), and false rejection rate (FRR). Comparative analysis shows consistent improvements of 3–6% across metrics, highlighting the system’s applicability to high-security authentication environments.
Keywords:	Generative AI, Deep Learning, Biometric Image Synthesis, Biometric Validation, Generative Adversarial Networks (GANs), Diffusion Models, Identity Verification, Anomaly Detection
Source:	Journal of Theoretical and Applied Information Technology 30^th September 2025 -- Vol. 103. No. 18-- 2025
Full Text

Title:	FedDNN-LDP: PRIVACY-PRESERVING FEDERATED RECOMMENDER SYSTEM USING DEEP NEURAL NETWORK AND LOCAL DIFFERENTIAL PRIVACY
Author:	THENMOZHI GANESAN , PALANISAMY VELLAIYAN
Abstract:	Federated recommender system utilizes federated learning to preserve user data in local and centralized system by training intermediate parameters instead of training raw user data. Currently, deep neural network gaining significant attention in recommender system due to its efficiency in processing in massive training samples and capturing intricate user-item interactions. However, global sharing of the entire user-item interaction in centralized network is prevented and limited for privacy concern. To address this gap, several integrated techniques utilizes the feature of pseudo-interaction of users and items in the neural network to recompense the missing values for each user and item, which adds random noise to the model and raise the privacy threatening in the network. This research proposes FedDNN-LDP, a novel federated deep neural recommendation system where centralized deep neural network is formed and preserved to enable the intricate user-item interactions on the server-end. On the client-end, to preserve the server from leaning user local data, pseudo-interaction is applied on the user data for data obfuscation. Furthermore, for privacy concern of local user-item interaction and obscure intermediate gradient parameters, integration of pseudo labeling and local differential privacy is exploited. Extensive experiments performed on two real world benchmark datasets namely MovieLens 100k and MovieLens 1M and experimental results show the outperformance of proposed model with existing competitive approaches ensuring privacy preservation by the following performance measures with the improvement: NDCG (15.47%), precision (30.06%) and recall (28.66%) for 100K dataset and NDCG (32.54%), precision (58.38%) and recall (40.86%) for 1M dataset. Further root mean squared error, mean absolute error, training and validation loss and hit ration are measured to exhibit the effectuality of the proposed model in terms of performance and privacy preservation.
Keywords:	Federated Recommender System, Deep Neural Network, Pseudo Labeling, Local Differential Privacy, Recommender System and Privacy Preservation.
Source:	Journal of Theoretical and Applied Information Technology 30^th September 2025 -- Vol. 103. No. 18-- 2025
Full Text

Title:	BAND SELECTION USING COEFFICIENT OF VARIATION AND BAND DENSITY RANKING IN HYPERSPECTRAL IMAGE
Author:	O. SUBHASH CHANDER GOUD, T. HITENDRA SARMA, C. SHOBA BINDU
Abstract:	In Hyperspectral Image (HSI) processing, one of the critical challenges is addressing Dimensionality Reduction through Feature Selection, especially given the high volume of spectral bands and often limited labeled data. This study introduces an innovative Band Subset Selection (BSS) technique that employs a Ranking-Based Approach to tackle this problem efficiently. The proposed approach is distinguished by its unsupervised nature, leveraging the fusion of two essential statistical measures: Coefficient of Variation (CV) and Band Density (BD). By synergistically combining these metrics, each band in the HSI dataset is analyzed, ranked, and subsequently filtered, allowing the model to identify an optimal subset of bands with the most relevant spectral information. This curated Band Subset (BS) method, termed CV-BDS-BS, is meticulously compared against an existing ranking procedure called SAM-SC (Spectral Angle Mapper with Spatial Coherence). Both methods undergo rigorous evaluation using state-of-the-art machine learning algorithms to ensure the efficiency, robustness, and reliability of the dimensionality reduction process. This integrated CV-BDS-BS methodology streamlines HSI data by reducing dimensionality and preserving essential spectral and spatial information.
Keywords:	Dimensionality Reduction, Feature Selection, Band Subset Selection, Coefficient of Variation, Band Ranking.
Source:	Journal of Theoretical and Applied Information Technology 30^th September 2025 -- Vol. 103. No. 18-- 2025
Full Text

Title:	LEARNING PATHWAYS OF HIGH-LEVEL ARGUMENTATION SKILLS IN ASYNCHRONOUS ONLINE DISCUSSION FORUM VIA EDUCATIONAL DATA MINING
Author:	SITI NAZLEEN ABDUL RABU , SITI KHADIJAH MOHAMAD , SITI NUR BAHIYAH KHALID
Abstract:	This study explores the learning pathways of high-level argumentation skills among undergraduate students participating in an asynchronous collaborative online discussion forum. Toulmin’s Model of Argumentation, with its six elements is employed to identify dominant types of argument contributions, while Peter and Wilson’s triads framework is used to assess the levels of argumentation. To uncover the pathways of argumentation skills, this study applies the data mining tool WEKA. Using a qualitative case study approach, the research combines content analysis and educational data mining to derive meaningful insights. A sample of 45 undergraduate students enrolled in an instructional multimedia course participated, with students grouped into nine small groups of five. The results show that while students were expected to explain their ideas with strong reasoning, most of their arguments remained basic. Evidence (36.5%) and Warrant (34.6%) were the most common contributions, as students often shared facts to support their opinions and attempted to connect those facts to their main points. However, more advanced argumentation skills, such as Rebuttal, i.e. questioning or challenging others’ views were rarely observed. Overall, 91.6% of student responses were classified as low-level, consisting mainly of simple opinions or explanations rather than deep, critical reasoning. Interestingly, not all groups performed equally. One group, G8_Skinner, demonstrated stronger, higher-level arguments. Students in this group participated more actively, contributed detailed explanations, and included supporting elements like Backing and Qualifiers, suggesting that active engagement and deeper reasoning go hand in hand. To explore how students can achieve stronger argumentation, a decision tree classifier was applied, identifying five key predictors of high-level performance: Add Post, Warrant, Backing, Qualifier, and Claim. The results show that students who posted more frequently, justified their reasoning, provided extra support, and used careful language were more likely to produce convincing arguments. However, the overall lack of rebuttals and limited critical engagement across most groups indicate a need for better instructional support to help students feel confident questioning ideas and building more complex arguments. These findings offer practical implications for educators, course designers, and educational technologists aiming to enhance argumentation, critical thinking, and student engagement in online learning environments.
Keywords:	Argumentation, Online Discussion Forum, Decision Tree Data Mining, Self-regulation, Reflection, Learning Analytics
Source:	Journal of Theoretical and Applied Information Technology 30^th September 2025 -- Vol. 103. No. 18-- 2025
Full Text

Title:	DIGITALIZATION OF HIGHER EDUCATION AS A TOOL FOR IMPROVING HUMANITARIAN SECURITY
Author:	OLEKSANDR HORBAN, MYKOLA STADNYK, LARYSA TARASIUK3, RUSLANA MARTYCH, SVITLANA VINTONIV-BAKHARIEVA, LEONID PANASIUK
Abstract:	Digitalization of the educational process provides access to quality knowledge and affects the development of thinking. The combination of the foundations of humanitarian security and digital technologies affects the consideration of the characteristics of modern society. The aim of the study is to determine the effectiveness of digitalization of higher education as a tool for strengthening humanitarian security. The research employed the following methods: analytical comparison, survey, observation, Likert scale, SWOT analysis, Student’s t-test. Digitalization of the educational process provided for the involvement of the Open edX application for the development of theoretical knowledge; the use of the Zoom platform to expand students’ pedagogical skills; the use of the Graasp platform to develop students’ creative abilities. It was established that the use of digital technologies in the educational process most contributed to the manifestation of such humanitarian security components as the ability to acquire academic knowledge (98%), the development of creative skills (97%). The results showed that after the study, students were able to achieve the development of critical thinking (98%), digital literacy (99%), emotional and psychological resilience (97%). The established indicators contribute to the development of humanitarian security. A SWOT analysis showed the advantage of strengths from involving digital technologies in the educational process to maintain humanitarian security. The practical significance of the article is the identified effective tools for implementing interactive learning for future teachers to improve humanitarian security. The research prospects are aimed at comparing traditional and digital approaches to education in order to determine their impact on the development of humanitarian security.
Keywords:	Digital Applications, Creative Opportunities, Self-Control, Educational Goals, Social Responsibility, Digitalization, Humanitarian Security.
Source:	Journal of Theoretical and Applied Information Technology 30^th September 2025 -- Vol. 103. No. 18-- 2025
Full Text

Title:	BLOCKCHAIN FOR SECURE AND TRANSPARENT VOTING SYSTEMS IN DEMOCRATIC ELECTIONS
Author:	D. LEELA DHARANI, MEDARAMETLA ANUSHA RANI, DR PRITI M BIHADE, DR O. RAMA DEVI, D VENKATA RAVI KUMAR, SIVAGANGA BADIPATI, DR S. SATHISH KUMAR, CHETLA CHANDRA MOHAN
Abstract:	Transparency For democratic processes to be fair, the mechanisms for voting must be kept honest and open. Menace plagues the conventional voting methods, involving physical presence that includes tampering of votes and fraud that destroys the voter's confidence. This work introduces a novel blockchain voting scheme which provides a decentralized, immutable, and secure environment to alleviate these problems. The solution combines biometric authentication for voters to confirm their identity, smart contracts for automatically counting votes, and blockchain to ensure no one can access and modify the voting records. A detailed simulation using artificial data ranging from 1,000 to 100,000 voters shows the efficiently provides vote integrity, authentication strength, scalability and latency. Experiment results have demonstrated that the proposed design can achieve 100% vote integrity, 99.8% authentication accuracy and much lower polling time than traditional e-voting systems. It also maintains real-time transparency using a blockchain ledger that can be used to audit voted ballots. This system provides a more secure, scalable, and transparent solution to trust-based voting methods. It can be used in future democratic elections to increase trust in the electoral process and decrease the possibility of fraud.
Keywords:	Blockchain, Voting System, Biometric Authentication, Smart Contracts, Transparency, Scalability
Source:	Journal of Theoretical and Applied Information Technology 30^th September 2025 -- Vol. 103. No. 18-- 2025
Full Text

Title:	MQTT-ENABLED IOT FRAMEWORK FOR EFFICIENT HEALTHCARE MONITORING AND PREDICTION SYSTEM
Author:	AFROZ PASHA, P S PRASAD, RIYAZULLA RAHMAN J, IMPA B H, SONIA MARIA D'SOUZA, BHAVANA A
Abstract:	For the clinically ailed subjects, the better enhancement, and opportunities in different applications given based on Internet of Things (IoT). Because of healthcare infrastructure, the residents of rural contains limited access for diagnostic and nursing services. As a result, when heart failure develops, people frequently fail to seek for help and use the services. An IoT provides an essential benefit in managing cardiac issues. This paper presents a novel stacked ensemble-based machine learning algorithm with cardio vascular (CV) monitoring model based on electrocardiogram (ECG) in corporate with IoT. The proposed framework involves the core steps namely data acquisition, data transmission based on IoT based Thingspeak cloud platform and CV prediction. Fetch the properties of ECG signal such as P, Q, R, S, and T then the pre-process the data using Pan Tomkins algorithm and it provide for CV prediction. For future health management, the level of age also forecasted. The hypertext transfer protocol (HTTP) and message queuing telemetry transport (MQTT) servers save the ECG data in the cloud. The error rate and ECG characteristics impacts are determined using Stacked Ensemble-based Machine Learning (SEML) algorithm. In an ECG monitoring device, utilize the PQRST regularity and achieve the acceptable outcomes. For CV patients, an effective cost with 95% prediction accuracy is accomplished via proposed methodology.
Keywords:	Heart Disease, Stacked Ensemble Machine Learning (SEML), Cardio Vascular, ECG signal and Thingspeak cloud platform.
Source:	Journal of Theoretical and Applied Information Technology 30^th September 2025 -- Vol. 103. No. 18-- 2025
Full Text

Title:	THE USE OF DIGITAL TECHNOLOGIES IN THE PROCESSES OF SOCIAL GROUPS CONSOLIDATION AS A TOOL FOR ENSURING STATE SECURITY
Author:	EDUARD PRYS
Abstract:	Digital technologies can serve as a pivotal element in strengthening social consolidation, which entails the cohesion of diverse social groups united by a single goal, particularly the enhancement of state security. The purpose of this study was to evaluate the synergistic influence of social consolidation and social integration within the digital environment, stemming from the interrelated actions of society and government concerning the state security level. The methods of correlation, variance and regression analysis are employed in the work. The study findings illuminated the correlation between indicators of social integration in the digital environment, social consolidation, institutional prerequisites, and state security. The regression analysis demonstrated, firstly, that the degree of social integration in the digital sphere is significantly contingent upon institutional prerequisites, with the model’s explanatory power reaching 58.57%. Secondly, the analysis revealed that the synergistic effect of social integration and social consolidation partially elucidates the level of state security, with an explanatory capacity of 31.92%. The results obtained may be instrumental in establishing a foundation for the formulation of policies aimed at harmonizing digital transformation with strengthening social stability.
Keywords:	Digital Technologies, Consolidation of Social Groups, State Security, Social Integration, E-Governance, Social Trust, Budget Transparency, Independence of Courts.
Source:	Journal of Theoretical and Applied Information Technology 30^th September 2025 -- Vol. 103. No. 18-- 2025
Full Text

Title:	ADDRESSING KEY MANAGEMENT AND DATA INTEGRITY IN IIOT BASED HEALTHCARE: A BLOCKCHAIN APPROACH USING FERNET ENCRYPTION AND MERKLE ROOT PROOF OF WORK
Author:	KUMAR M P , AKILA A
Abstract:	A blockchain based healthcare system in the Industrial Internet of Things (IIoT) improves transparency and interoperability by decentralizing patient data access and storage. It ensures tamper-proof records by enabling secure and seamless sharing of healthcare information among stakeholders and connected devices. However, IIoT devices are vulnerable to security issues like unauthorized attacks due to improper key management, which makes it easy for attackers to steal data. To overcome this issue, the Fernet Encryption Algorithm with Merkle Root Proof of Work (FEA-MRPoW) is proposed in blockchain-based healthcare systems for IIoT to increase security by generating strong data encryption which prevents unauthorized access and data confidentiality. The structure of Merkle root provides integrity of healthcare records whereas PoW protect against tampering for block validation. Therefore, this integration provides robust performance and ensures the safety of healthcare data in IIoT environments. The proposed FEA-MRPoW achieves a less encryption time of 0.298s when the number of attributes is 10 compared to existing methods like the Diagonal Digital Signature Algorithm (DDSA) with Merkle Patricia Hash Trie (MPHT).
Keywords:	Blockchain, Fernet Encryption Algorithm with Merkle Root Proof of Work, Healthcare, Industrial Internet of Things, Unauthorized Access.
Source:	Journal of Theoretical and Applied Information Technology 30^th September 2025 -- Vol. 103. No. 18-- 2025
Full Text

Title:	NEURAL NETWORK-BASED ENSEMBLE APPROACH TO DETECT INFORMED ATTACKS FOR MULTIPLE TARGETS IN RECOMMENDER SYSTEMS
Author:	ASHISH KUMAR , YUDHVIR SINGH
Abstract:	Recommender systems (RSs) based on collaborative filtering (CF) are frequently used to deliver personalized services. However, profile injection attacks can easily affect them. As a result, attackers can easily manipulate the outcomes of these RSs. Informed attacks, which are also a type of profile injection attack, are very challenging to identify due to their high resemblance to genuine user profiles. Very limited research has been carried out to identify informed attack profiles, so there is a significant research gap to propose a technique to identify these profiles with good accuracy. In our experiment, we proposed a new data partition scheme for better training and testing of machine learning models. We injected informed attacks to promote and demote a specific item and a set of 10 items in the MovieLens dataset, and we proposed a novel neural network-based ensemble approach. Its performance is evaluated based on accuracy, precision, and recall by comparing it with that of a classical voting-based ensemble model (CVBEM), along with other supervised machine learning models in the detection of informed attacks. Robustness of performance is ensured by using k-fold cross-validation. We conducted our experiment in 24 attack scenarios with varying types of attacks, intentions, target item sizes, and attack sizes. Our study found that the proposed model's accuracy outperforms the other models' accuracy by a good margin of nearly 4% in most of the test scenarios. The CVBEM comes out as the second-best performer among all. The proposed model performs better not only in predicting biased users but is also more stable compared to traditional machine learning models.
Keywords:	Informed Attacks, Ensemble Model, Recommender System, Probe Attacks, Power User Attacks, Neural Network
Source:	Journal of Theoretical and Applied Information Technology 30^th September 2025 -- Vol. 103. No. 18-- 2025
Full Text

Title:	BIG DATA AND ARTIFICIAL INTELLIGENCE REVOLUTIONIZING FINANCIAL FRAUD DETECTION SYSTEMS
Author:	CHALLAPALLI SUJANA, A. SASI HIMABINDU, DR. DIVVELA SRINIVASA RAO, SASIKALA RASAMSETTY, ARUNKUMAR M S, P. MUTHUKUMAR, SATHISH KUMAR SHANMUGAM
Abstract:	Fraud detection is a significant challenge when it comes to detecting financial fraud, as the activities of financial fraud occur at a high frequency. The model of Big Data analytics hybrid with Artificial Intelligence, which comprises a convolutional neural network (CNN) and a long short-term memory (LSTM) network, is presented in the current paper. The aim is to enhance the detection rate and minimize the rates of false positives and false negatives, especially when fraudulent transactions are identified. The predicted system is implemented in the form of feature extractors, utilizing CNNs after LSTMs that model temporal dependence on transaction data. The synthetic data set on which it was tested has been designed to emulate the real-life application of the model in making financial transactions, and it performed better than the conventional machine-learning algorithms in Random Forest, SVM, and Gradient Boosting (accuracy, 96.2%; precision, 95.2%; recall, 92.6%). The findings indicate that our hybrid CNN-LSTM solution is feasible for carrying out fraud detection with a relatively low false positive rate, which is quite significant in preventing customer inconvenience. The implications of this model are powerful, as it provides the financial industry with a real-time, scalable, and efficient solution to prevent fraud, streamline business procedures, and foster customer trust in the industry.
Keywords:	Financial Fraud Detection, Big Data Analytics, Artificial Intelligence, Convolutional Neural Network, Long Short-Term Memory, Fraud Classification
Source:	Journal of Theoretical and Applied Information Technology 30^th September 2025 -- Vol. 103. No. 18-- 2025
Full Text

Title:	IMPROVED ATTENTION-BASED RCNN SEGMENTATION AND ENSEMBLE CLASSIFIER FOR LUNG CANCER CLASSIFICATION AND SEVERITY LEVEL ASSESSMENT USING CT IMAGE
Author:	A NAGA KALYANI , VIJAYA KUMAR V
Abstract:	Lung cancer (LC) is a horrible ailment that impacts many countries across the world, and early identification remains difficult. Oncologists analyze the cancer using blood specimens and CT images, which consumes time and requires additional human work. To minimize mortality, an automated approach for identifying and assessing lung cancers should be developed. To address this, a new approach is proposed in this paper for LC classification (LCC), which initially employs an Averaged Gaussian-Median Filtering (AG-MF) model for pre-processing the CT image. In the next phase, a segmentation is performed with an Improved Attention Layer-Based Mask RCNN (IAL-MRCNN). Subsequently, features, like deep features, Entropy features, statistical and color features are derived. Then, detection and severity classification are done by utilizing an ensemble classifier combining Improved AlexNet, Bi-LSTM, and SqueezeNet to accurately classify lung tumors and assess their severity. From the analysis, the high precision proves the classification performance, as the improved AlexNet achieved a higher value of 0.956 at 90% of training data.
Keywords:	Lung cancer; Averaged Gaussian-Median Filtering; IAL-MRCNN; SqueezeNet; Improved AlexNet.
Source:	Journal of Theoretical and Applied Information Technology 30^th September 2025 -- Vol. 103. No. 18-- 2025
Full Text