		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
January 2026 | Vol. 104 No.2

Title:	ENHANCING TASK RELATED KNOWLEDGE SHARING IN AGILE VIRTUAL TEAMS: FROM INFLUENCING FACTORS TO PRACTICAL GUIDELINES – KSAVT FRAMEWORK
Author:	ZOHAIB AHMED, ZULKEFLI MANSOR, KAMSURIAH AHMAD, MUHAMMAD ZAFARULLAH
Abstract:	Task-relevant knowledge sharing (TaRK) is an essential process in maintaining alignment, collaboration, and learning within geographically dispersed teams in distributed agile environments. Established on the existing Knowledge Sharing for Agile Virtual teams (KSAVT) Framework, this study clarifies the role of task-level practices in ensuring effective knowledge sharing and improving collective performance. Based on a qualitative multiple case study of three international software organizations and eight practitioners, the research validates seven task-related influencing factors (IFs) and offers eight practical guidelines that put these influencing factors into practice in real-world agile settings. The results suggest that sustaining knowledge sharing in distributed agile environments depends on open communication, workload sharing, repeated reflection, and relevant filtering. In theory, the study adds to the KSAVT Framework by showing how efficient task related knowledge sharing can convert an individual’s capacity of doing a task into team’s capability. In practice, it gives agile teams clear steps to make task related knowledge sharing a part of their daily routine, which helps them become more flexible, learn faster, and work better together. Consequently, task-oriented knowledge sharing becomes an active, iterative process that turns everyday work into learning loops, promoting adaptability and sustained team performance.
Keywords:	Agile Virtual Teams (Avts), Task Related Knowledge (Tark), Knowledge Sharing, Influencing Factors (Ifs), Industrial Case Study, Global Software Development (GSD), Knowledge Sharing For Agile Virtual Teams (KSAVT) Framework.
DOI:	https://doi.org/10.5281/zenodo.18453956
Source:	Journal of Theoretical and Applied Information Technology 31^st January 2026 -- Vol. 104. No. 2-- 2026
Full Text

Title:	EFFECTIVE CBIR BASED ON HYBRID IMAGE FEATURES AND MULTILEVEL APPROACH
Author:	V. ARCHANA REDDY, V. VIJAYA KUMAR
Abstract:	The instantaneous search and retrieval of the most relevant images to a specific query is one of the significant applications of image processing. The process of retrieval of images using image contents is widely known as content base image retrieval (CBIR). The image features extracted from the local windows of 33 resulted good results in CBIR. However, the micro windows of 22 derived the texton and motif features and played a dominant role in CBIR. This paper segmented the 33 window into two windows namely cross and diagonal windows. Four directional motifs of 13 are extracted from each of the cross and diagonal segments. The motif features are derived using Rule based Directional motif (RDM) to address the ambiguity issues. This paper transformed the motif indexes derived on a 1x3 triangular window to a 33 window and derived eight triangular RDM local units. The local features are extracted by integrating the features extracted from the eight different images. The extracted features hold directional, textures, patterns, edge properties derived from the 33 and triangular windows. The results indicate the superiority of the proposed methods.
Keywords:	33 Window, 13 Triangular Window, Directional Motif;
DOI:	https://doi.org/10.5281/zenodo.18453998
Source:	Journal of Theoretical and Applied Information Technology 31^st January 2026 -- Vol. 104. No. 2-- 2026
Full Text

Title:	ADVANCEMENTS IN QUANTUM COMPUTING WITH THE DEVELOPMENT OF TOPOLOGICAL QUBITS FOR SCALABLE AND FAULT-TOLERANT SYSTEMS
Author:	DR. KONGARA SRINIVASA RAO,DR K SREERAMA MURTHY, CHETLA CHANDRA MOHAN,G. VASUNDHARA DEVI, YADAVALLI SURESH KUMAR,VEERAMACHINENI VENKATA RAGHU, MANASA BANDLAMUDI, UDAYA LAKSHMI SAKALA
Abstract:	Quantum computing can help solve complex problems classical computers cannot solve. Current quantum systems, especially when using conventional qubits sensitive to environmental noise, face key scalability and fault tolerance hurdles. This has been identified as a promising path and is, in principle, fault-tolerant, but it still suffers from control and scalability issues. This work presents a new paradigm of quantum computation based on a hybrid quantum system combining topological qubits for error correction and traditional qubits for computing. The final hybrid system can reduce the bit-flip error probability from 0.05 to 0.02 and the phase-flip error probability from 0.08 to 0.01. The fidelity of quantum states after gate operations reached 0.92, surpassing both that of conventional (0.85) and topological (0.90) qubits. Moreover, this approach achieved a strength of interaction (0.90) and operation time (0.01 ms) higher than conventional qubits. Hybrid architecture significantly improves quantum systems' efficiency, scalability, and fault tolerance. Even with these advancements, controlling topological qubits and measuring the braiding of anyons are still challenging. Soon, research will concentrate on overcoming these technical challenges and paving the way for scaling up the hybrid system to more complex quantum computations.
Keywords:	Quantum Computing, Topological Qubits, Majorana Fermions, Quantum Error Correction, Scalable Systems, Fault Tolerance, Hybrid Quantum Systems
DOI:	https://doi.org/10.5281/zenodo.18454014
Source:	Journal of Theoretical and Applied Information Technology 31^st January 2026 -- Vol. 104. No. 2-- 2026
Full Text

Title:	AI-DRIVEN EXPLICIT AND IMPLICIT INFORMATION COLLECTION IN REQUIREMENTS ENGINEERING: CLASSIFICATION AND KNOWLEDGE REPRESENTATION
Author:	AHMAD MURAD, KHALIL JAMOUS, IHAB ATIEH, AMJAD HUDAIB, NADIM OBEID, MARWAN AL-TAWIL
Abstract:	The evolving complexity of software systems demands smarter Requirements Engineering (RE) methods that can uncover both explicit and implicit user needs. Traditional RE techniques often miss latent contextual information, leading to incomplete or suboptimal system specifications. This paper introduces an AI-enhanced framework that integrates Natural Language Processing (NLP), Machine Learning (ML), and Knowledge Graphs (KGs) to intelligently extract, and classify requirements. Explicit requirements are derived from structured documentation using BERT and TF-IDF, while implicit requirements are inferred from behavioral patterns and unstructured data. Class imbalance is mitigated using SMOTE and ADASYN techniques, and deep learning via BiLSTM captures bidirectional semantic dependencies. The PROMISE dataset is used to train and evaluate the model, with results benchmarked using accuracy, F1-score, and ontology completeness. The resulting knowledge graph enables contextual traceability. Our framework demonstrates improved requirement visibility, and semantic enrichment, moving RE toward a more intelligent, context-aware discipline.
Keywords:	Requirements Engineering RE, Explicit and Implicit Requirements, Knowledge Graphs KG, Natural Language Processing NLP, Machine Learning ML, AI-Driven Requirements Analysis
DOI:	https://doi.org/10.5281/zenodo.18454041
Source:	Journal of Theoretical and Applied Information Technology 31^st January 2026 -- Vol. 104. No. 2-- 2026
Full Text

Title:	COURSE FORGE AI IN COMPUTERS: AN AGENTIC, MODULAR, AND EXPLAINABLE EDUCATIONAL ASSISTANT USING LLMS AND RAG
Author:	LENIN MARKSIA U, JEYASHANTHI J, JEGADEESH A, KALIAPPAN M, MARIAPPAN E, ANGEL HEPZIBAH R,RAMANTH M,MEGA G
Abstract:	The demand for personalized, interactives learning remains a core challenge in modern education. This paper presents CourseForge AI, a modular and explainable academic assistant designed using Large Language Models (LLMs), autonomous agents, and Retrieval-Augmented Generation (RAG). The system includes specialized agents for curriculum-aligned Question and Answer, note generation, flashcards, tutoring, and coding assistance and all powered by semantic retrieval over academic content. Unlike static chatbots, Course Forge AI exhibits agentic reasoning and decision-making, allowing it to dynamically retrieve and contextualize content in response to user input. By leveraging vector-based semantic search model includes LanceDB and explainable agent workflows through Phidata, the system ensures that its outputs are grounded, interpretable, and relevant to syllabus-specific learning. Experimental validation shows strong performance in user engagement, relevance, and accuracy. This work contributes a scalable blueprint for AI-assisted education that is both intelligent and trustworthy
Keywords:	Agentic AI, AI Assisted learning, Curriculum-Aligned Learning, Large Language Models (LLMs), Retrieval-Augmented Generation (RAG)
DOI:	https://doi.org/10.5281/zenodo.18454059
Source:	Journal of Theoretical and Applied Information Technology 31^st January 2026 -- Vol. 104. No. 2-- 2026
Full Text

Title:	MULTI-ASPECT SENTIMENT ANALYSIS IN AMAZON REVIEWS USING GMM-ENHANCED TEXTCAPS WITH PROBABILISTIC CAPSULE ROUTING
Author:	SRI SOWMIYA.S , Dr. L. SANKARI
Abstract:	Online shopping enables consumers to purchase goods through e-commerce platforms, where customer reviews significantly influence buying decisions and vendor reputation. Sentiment analysis of these reviews facilitates the decoding of public opinion, offering actionable insights for product improvement, marketing personalization, and an enhanced customer experience. The reviews often contain mixed sentiments, sarcasm, and abrupt polarity changes, which traditional models tend to oversimplify, thereby losing sentiment flow and interpretability. This study proposes GMM-Enhanced TextCaps, an advanced model addressing these challenges by employing Byte-Pair Encoding for subword-level sentiment extraction, capsule networks for granular aspect representation, and Gaussian Mixture Models for probabilistic sentiment routing. The model processes enriched token embeddings through contextual transformer encoders, sarcasm and negation adjustments, and emotion-weighted attention mechanisms, forming interpretable capsules that yield confident sentiment predictions. Experiments utilize the Clothing, Shoes, and Jewelry category from the Amazon Reviews ’23 dataset, comprising 451,478 reviews divided into 80% training, 10% validation, and 10% testing splits. The model attains a balanced accuracy of 72.51% and a G-Mean of 72.45%, demonstrating robust performance in capturing nuanced sentiment dynamics while effectively maintaining class balance.
Keywords:	Aspect-Based Sentiment Analysis, Capsule Networks, Gaussian Mixture Models, Sentiment Interpretation, Multimodal Sentiment Analysis, Transformer Encoding
DOI:	https://doi.org/10.5281/zenodo.18454074
Source:	Journal of Theoretical and Applied Information Technology 31^st January 2026 -- Vol. 104. No. 2-- 2026
Full Text

Title:	AI-ADAPTIVE POST-QUANTUM CRYPTOGRAPHY FOR SECURE AND FRAUD-RESILIENT DIGITAL PAYMENTS
Author:	BALAKRISHNAN S , UPPALAPATI NAGA RATNA KUMARI, DR.G.GOKUL KUMARI , DR. LAMA SAMEER KHOSHAIM , S. RAMYA , ANANTHA RAO GOTTIMUKKALA
Abstract:	The threat of quantum computing to classical cryptographic infrastructures is steadily increasing alongside more complex and multi-modal payment fraud posing a severe security gap within current financial infrastructures. Current payment stacks, which are based on TLS and have fixed classical cryptography and tabular fraud detectors, are inadequate to secure high volume digital transactions against both quantum decryption attacks as well as coordinated fraudulent behaviour. The purpose of the study was to create and test an AI-based framework that can optimally improve fraud detection and provide adaptive post-quantum cryptographic protection in real-time. To satisfy this goal, we developed an experimental system that merged a heterogeneous temporal graph neural network (HTGNN) to multimodal fraud analysis with a reinforcement-learning PQC orchestrator that can be used to choose NIST-standard ML-KEM and ML-DSA parameter sets based on transaction-level risk. Multimodal datasets (tabular features, device metadata, text fields, and relational graphs) were publicly available and converted into a fused format and tested in adversarial, latency, and throughput-controlled experiments. PQC-enabled OpenSSL/liboqs was used to implement the crypto layer to measure handshake performance in the real world. Findings indicate that the proposed model enhanced the AUPR of fraud-detection (baseline) of 0.35 to 0.58 and increased ROC-AUC to 0.966, which was better than the classical and hybrid systems. The adaptive PQC algorithm decreased the handshake latency to 68 ms, which was significantly lower than setups of PQC (90-120 ms). Robustness tests showed a smaller degradation in recall (−9) and a cipher-downgrade success rate of less than 1% which depicts a better resilience to the threat vectors. Altogether, the research results show that a combination of multimodal GNN-based fraud detection and AI-adaptive PQC is a viable and quantum-resilient security architecture of digital payment systems in the next generation.
Keywords:	Digital Payments Fraud, Post-Quantum Cryptography, Adaptive Security, Graph Neural Networks, Reinforcement Learning.
DOI:	https://doi.org/10.5281/zenodo.18454191
Source:	Journal of Theoretical and Applied Information Technology 31^st January 2026 -- Vol. 104. No. 2-- 2026
Full Text

Title:	FEDERATED LEARNING FOR PRIVACY PRESERVING MACHINE LEARNING IN INTERNET OF THINGS IOT NETWORKS CHALLENGES SOLUTIONS AND FUTURE DIRECTIONS
Author:	P. SURIYA, DR. P SUMITHABHASHINI, EERLA RAJESH, KALADI GOVINDARAJU, VUPPULOORI RAVI SEKHARA REDDY, ANANTHA RAO GOTTIMUKKALA, SRIJA GUNDAPANENI, K SWETHA
Abstract:	The explosive growth of Internet of Things (IoT) devices has led to large amounts of sensitive data production, causing privacy concerns for traditional machine learning models. One such approach is Federated Learning (FL), which enables the training of models in a decentralized manner, keeping the data on devices. Issues like communication overhead, privacy preservation, and scalability are still present, especially in resource-constrained IoT scenarios. We propose a new framework of FL in the context of IoT networks with mechanical privacy protection like differential privacy and homomorphic encryption, such as model compression and dynamic client selection. The framework was tested against standard IoT datasets and showed a better performance of 88.2% model accuracy than FedAvg (84.6%), FedProx (85.9%) and HeteroFL (83.4%). The results showed that overhead communication was reduced by as much as 15% while our system could efficiently scale to 1000 clients with a negligible increase in the per-round training time. In summary, the presented framework achieves high accuracy, privacy, and efficiency and is highly scalable for large IoT networks, but it still suffers from limitations such as resource constraints and parameter tuning for privacy preservation. In future work, we will work on making our framework even more robust, integrating asynchronous learning, and testing it in real-life IoT scenarios to achieve better scalability and applicability.
Keywords:	Federated Learning, Internet of Things, Privacy Preservation, Communication Efficiency, Model Accuracy
DOI:	https://doi.org/10.5281/zenodo.18454221
Source:	Journal of Theoretical and Applied Information Technology 31^st January 2026 -- Vol. 104. No. 2-- 2026
Full Text

Title:	A DUAL-BACKBONE SWIN-TRANSFORMER AND EFFICIENTNET FRAMEWORK FOR ACCURATE ALZHEIMER’S DISEASE STAGE PREDICTION
Author:	M K V ANVESH, PRAJNA BODAPATI
Abstract:	Alzheimers disease (AD) is a serious health problem throughout the world that gradually damages memory and thinking abilities in human beings. Identifying the precise stage of Alzheimer’s disease (AD) is a complex task for clinicians, as the disorder progresses through stages such as Non-Demented, Very Mild Demented, Mild Demented, and Moderate Demented. However, current diagnostic tools often struggle to correctly sort patients into these groups, these kind of problems frequently lead to wrong or late diagnoses. To address these challenges, this study proposes a dual-backbone deep learning framework that integrates Swin Transformer with EfficientNet-B7. The model was trained using the Alzheimer's Synthesized Dataset, which provides 8,000 pseudoRGB MRI images (roughly 2,000 images per disease stage). All images were prepared by being resized to pixels, put through various augmentations (changes), and then normalized. The model itself uses two specialized components: the Swin Transformer to pick up large-scale brain patterns, and EfficientNet-B7 to focus on fine, small local details. The special features from the two models are merged and passed to a single classifier, which makes the final, accurate prediction across the four disease classes. The model showed best results, achieved an accuracy of 99% and strong F1-scores after being trained with an 85% training and 15% validation data split. Further, Grad-CAM heatmaps are used to provide clear explanations and confirm that the results are reliable. This combination of high performance and clarity makes the overall system a promising and dependable method for quick Alzheimers diagnosis.
Keywords:	Alzheimers Disease Classification, Swin Transformer, EfficientNet-B7, pseudoRGB MRI, Dual-Backbone Deep Learning, Grad-CAM.
DOI:	https://doi.org/10.5281/zenodo.18454233
Source:	Journal of Theoretical and Applied Information Technology 31^st January 2026 -- Vol. 104. No. 2-- 2026
Full Text

Title:	CONSTRAINT-AUGMENTED NEURAL NETWORKS FOR SHAKESPEAREAN VERSE: INTEGRATING PROSODIC FIDELITY WITH TRANSFORMER MODELS
Author:	MEHULKUMAR H KANTARIA , SAHANA EDWIN , VISHAL NAMIREDDY, S. KRISHNAKUMARI , JAKKAPU NAGALAKSHMIDEVI , DR GANTA JACOB VICTOR
Abstract:	This research proposes a methodology for augmenting prosodic constraints to enhance the creation of Shakespearean poetry in neural models of language. Contemporary transformer-based generators are proficient in fluency but often do not satisfy the stringent metrical and rhyming standards of formal poetry. To fill this gap, the proposed technique uses specialized prosody embeddings as well as prosody-aware attention mechanisms to combine explicit prosodic information such syllable counts, stress patterns, along with rhyme suffixes. The model is guided toward simultaneous optimization of linguistic quality as well as prosodic integrity by a composite training goal that combines cross-entropy loss with distinct meter and rhyme penalties and optional reinforcement learning. Experiments on a prosodically annotated Shakespearean corpus demonstrate substantial gains in meter accuracy, rhyme correctness, and overall stylistic adherence compared with standard GPT-based baselines, without sacrificing coherence or fluency. The results highlight the effectiveness of embedding phonological structure directly into neural architectures and offer a scalable approach for controllable poetic and stylistically constrained text generation.
Keywords:	Prosodic Constraint Augmentation, Shakespearean Verse Generation, Neural Language Models, Meter and Rhyme Modelling, Controllable Text Generation
DOI:	https://doi.org/10.5281/zenodo.18454285
Source:	Journal of Theoretical and Applied Information Technology 31^st January 2026 -- Vol. 104. No. 2-- 2026
Full Text

Title:	FINTECH AS A FACTOR OF ASYMMETRIC TRANSFORMATION OF THE BANKING SECTOR: EUROPEAN EXPERIENCE AND POTENTIAL FOR UKRAINE
Author:	CRISTINA GABRIELA COSMULESE, ARTUR ZHAVORONOK, ROMAN GRESHKO, NATALIA OSTROVSKA, VIOLETTA KHARABARA, VIKTORIIA TKACH
Abstract:	In the current conditions of digital transformation of the financial sector, FinTech is one of the key drivers of structural changes in banking. The COVID-19 pandemic, the full-scale war in Ukraine, and global digitalization processes have significantly accelerated the introduction of contactless payment technologies, mobile banking, and remote financial services. Special scientific attention is required to analyze the impact of FinTech on the banking systems of developing countries, which simultaneously strive to catch up with digital leaders and face institutional, economic, and infrastructural constraints. Ukraine, being in conditions of military instability, demonstrates unexpectedly high rates of digitalization of banking services, which actualizes the need for its scientific understanding in a comparative international context. The purpose of the study is to analyze the main trends in the development of financial technologies in the banking sector of different countries, in particular, to compare digital transformation in developed and developing countries, to identify success factors, development barriers, and the potential for integrating innovations into the banking sector. The study uses quantitative and comparative approaches. Statistical indicators characterizing access to financial services and the level of digitalization of banking activities are analyzed, in particular, the number of bank branches and ATMs, the volume of mobile and Internet banking transactions, the number of transactions per capita and the share of digital payments in GDP using the example of 11 European countries. Methods of comparative analysis, data visualization and interpretation of digital trends in dynamics are used. The results of the study indicate the presence of pronounced unevenness and asymmetry of the digital transformation of the banking sector under the influence of FinTech. In developed countries (Estonia, Poland, Turkey), FinTech is actually transforming into the dominant banking service channel, which is accompanied by a reduction in physical banking infrastructure, a rapid increase in the number of digital transactions and a high share of non-cash payments in GDP. In contrast, in developing countries (Albania, Kosovo, Bosnia and Herzegovina, Moldova), digital financial services are at the stage of formation and are developing in parallel with the physical banking infrastructure, which indicates the catching-up nature of digitalization and limited substitutability of traditional service channels. For Ukraine, despite the lack of complete statistical data in the sample, a specific model of forced digital transformation is characteristic, in which the reduction of physical banking infrastructure is combined with the intensive growth of remote banking services, in particular mobile banking and non-cash payments. This allows to attribute Ukraine to the group of countries with a high potential for accelerated FinTech integration, provided that the institutional and regulatory environment is stabilized. The value of the study lies in identifying and systematizing models of asymmetric FinTech transformation of the banking sector in European countries with different levels of economic development, which allows to explain the absence of a universal trajectory of digitalization. The practical significance of the results obtained lies in the possibility of using them to form strategies for the digital transformation of the banking sector, develop financial inclusion policies, and adapt best European FinTech practices for countries in catching up, in particular for Ukraine.
Keywords:	ICT Sector; FinTech; banking sector; digital transformation; developing countries; IMF Financial Access Survey; digital transactions; cashless payments; banking services market; national economy; Ukraine; Europe.
DOI:	https://doi.org/10.5281/zenodo.18454300
Source:	Journal of Theoretical and Applied Information Technology 31^st January 2026 -- Vol. 104. No. 2-- 2026
Full Text

Title:	BALANCING ACCURACY AND ROBUSTNESS IN INTRUSION DETECTION: AN ENSEMBLE DEEP LEARNING APPROACH
Author:	U.V. RAMESH, RAMESH KOTHAPALLI, CH BHANU PRAKASH, P. SRILATHA4,SANDA SRI HARSHA, ANANTHA RAO GOTTIMUKKALA
Abstract:	Intrusion Detection Systems increasingly face adaptive, multimodal cyberattacks, yet most of the current systems are highly accurate in clean conditions and inadequately robust to adversarial perturbations, creating a major robustness gap. This challenge is necessary to ensure the current network settings where the attacks are modified at the structural, time, and content levels. The purpose of this study was to design and test a sound multimodal intrusion detection system that is able to meet accuracy and adversarial resilience. Our experimental study was based on four publicly available datasets of IDS, which combined the flow features, packet payloads, graph-structured interactions with the host telemetry into a single architecture. The model was called Robust Multimodal Graph-Transformer Ensemble (RMGTE) and contrastive alignment, adversarial training, Mixture-of-Experts routing and randomized smoothing were used as its additions to improve the robustness and was evaluated against RF, GNN and hybrid models. Results indicate that RMGTE attained 96.1% accuracy, 0.93 macro-F1, and 0.61 adversarial accuracy, which is significantly higher than baselines; furthermore, it had a certified robustness radius of 0.48, and high cross-dataset generalization (0.87 macro-F1 on BoT-IoT holdout). These results indicate that multimodal fusion and robustness-aware ensemble design makes a substantial contribution to the performance and reliability of IDS, which provides a promising future path to resilient next-generation cybersecurity systems.
Keywords:	Intrusion Detection, Multimodal Deep Learning, Graph Transformer, Adversarial Robustness, Ensemble Learning
DOI:	https://doi.org/10.5281/zenodo.18454317
Source:	Journal of Theoretical and Applied Information Technology 31^st January 2026 -- Vol. 104. No. 2-- 2026
Full Text

Title:	A COMPUTATIONALLY-STRUCTURED FRAMEWORK FOR GENERALIZABLE AND EXPLAINABLE AI IN INJECTION MOLDING: INTEGRATING DATA STANDARDIZATION, TRANSFER LEARNING, AND EDGE DEPLOYMENT
Author:	MOHAMMED HDID , HAKIM JEBARI , HASSAN SAMADI
Abstract:	The emergence of Artificial Intelligence (AI) and Machine Learning (ML) is now the key element of modern manufacturing, but the majority of the models applied in injection molding are domain-specific and cannot be generalized to other machines, materials, or process conditions. The given paper deals with this computing problem by developing a structured literature review devoted to the issue of the generalization and transferability of industrial AI systems. Based on this discussion, a conceptual framework of 5 steps is put forward to formalize the way in which cross-domain robustness may be attained based on fundamental computing systems like data normalization, feature relevance control, regularization, domain adaptation, and explainable modeling. The stages are all based on the latest developments in AI, the introduction of digital-twin, and the integration of hybrid edge-cloud architecture. The framework suggested opens up a hypothetical basis of constructing domain resistant and intelligible artificial pipelines, that links machine particular learning with scalable and transferable intelligence. This effort to coordinate the data of industrial processes with the principles of reproducibility and generalization in computing leads to the further development of both AI methodology and practical use of smart manufacturing in the direction of the reliable and cross-domain automatization.
Keywords:	Plastic Injection Molding, Machine Learning (ML), Artificial Intelligence (AI), Process Optimization, Production Efficiency, Feature Normalization, Defect Detection and Prediction, Explainable AI (XAI), Predictive Maintenance, Digital Twin.
DOI:	https://doi.org/10.5281/zenodo.18454333
Source:	Journal of Theoretical and Applied Information Technology 31^st January 2026 -- Vol. 104. No. 2-- 2026
Full Text

Title:	HYBRID MACHINE LEARNING MODELS FOR INTEGRATED URBAN AIR QUALITY AND WEATHER FORECASTING - A DECISION-SUPPORT FRAMEWORK FOR SUSTAINABLE CITY PLANNING
Author:	P. NAGAMANI, GORANTLA NIPUN, AYINALA NAGA SAI, G KIRAN KUMAR, D KARTHIK PHANI VARMA, DR. K. SWATHI, DR. N. NEELIMA, DR.T.V.SAI KRISHNA
Abstract:	Rapid urbanization has intensified challenges related to air pollution and extreme weather, directly impacting public health and sustainable city planning. This study presents a hybrid machine learning framework for integrated air quality and weather forecasting using environmental, meteorological, and urban data. The model combines regression, classification, and time-series prediction techniques to capture complex interactions. Experimental results show improved air quality prediction (RMSE = 3.45) and weather forecasting accuracy (93.6%), demonstrating the framework’s effectiveness as a decision-support tool for climate-resilient and sustainable urban development.
Keywords:	Machine Learning, Air Quality Prediction, Weather Prediction, Urban Planning, Hybrid Model
DOI:	https://doi.org/10.5281/zenodo.18454343
Source:	Journal of Theoretical and Applied Information Technology 31^st January 2026 -- Vol. 104. No. 2-- 2026
Full Text

Title:	IMPROVING ACCURACY IN FRAUD DETECTION IN PUBLIC COMPANIES FINANCIAL REPORTS USING NATURAL LANGUAGE PROCESSING
Author:	ANGELA KURNIAWAN, LAUDYA NATALIE YUDHA, ARMANTO WITJAKSONO
Abstract:	This study develops and evaluates a Natural Language Processing (NLP) model for detecting financial statement fraud in Indonesian public companies through linguistic analysis of annual reports. We analyzed 72 financial reports (20 fraudulent, 52 non-fraudulent) from companies listed on the Indonesia Stock Exchange using TF-IDF-based text representation combined with linguistic complexity features. Machine learning classifiers (Logistic Regression, Random Forest) were evaluated using 5-fold cross- validation. The NLP model achieved 77.3% accuracy, significantly outperforming traditional rule-based audit method (40.9%) by 88.9%. Contrary to theoretical expectations, fraudulent reports used simpler language (ambiguity score: 0.69 vs. 0.98) but exhibited significantly higher hedging language usage (p=0.033). TF-IDF features dominated model performance (99.9% contribution). This study is the first to systematically apply NLP-based fraud detection to Indonesian financial statements in a bilingual reporting context, revealing context-specific linguistic patterns that differ from Western fraud literature.
Keywords:	Natural Language Processing, Fraud Detection, Public Companies Financial Reports, Artificial Intelligence, Manipulatio006E
DOI:	https://doi.org/10.5281/zenodo.18454367
Source:	Journal of Theoretical and Applied Information Technology 31^st January 2026 -- Vol. 104. No. 2-- 2026
Full Text

Title:	PREDICTING INTERCONNECTED CLIMATE IMPACTS USING THE COX PROPORTIONAL HAZARDS MODEL AND XAI
Author:	Y.YESUJYOTHI, DR.SUBHANI SHAIK, DR.YALLA VENKAT
Abstract:	Climate change has systemic and cumulative effects on air quality, water availability, food production, and waste management, yet most current predictive studies examine each area separately and provide little guidance on how to interpret them to inform policy decisions. This paper interprets climate change as a systemic risk over time and proposes a feature that integrates the Cox proportional hazards model with explainable artificial intelligence (XAI) to measure and explain cascading climate effects. NASA, NOAA, and environmental monitoring agencies, as well as peer-reviewed studies, were combined to create multisource longitudinal datasets covering the 1980-2024 timeframe and including atmospheric, hydrological, agricultural, and waste-related variables. The Cox model can be used to estimate risks in time-dependent settings where all sectors are mutually dependent, whereas Shapley Additive Explanations (SHAP) can provide clear explanations of the effect of each variable. The experimental evidence shows that the suggested framework attains a concordance index of more than 0.82, which is better than traditional regression-based models for predicting changes in climate risk. Temperate variations and CO2 release become the predominant causes of agricultural and water-induced risks, with PM 2.5 concentration exerting the greatest influence on the escalation of health risk. In contrast to earlier models based on the concepts of staticity and black-boxing, the proposed method provides interpretable, testable, and policy-relevant insights into the dynamics of climate risks. The findings underscore the evidence-based approaches to climate adaptation and provide a scalable, explainable basis for integrated climate resilience planning, especially in susceptible and rapidly urbanizing areas.
Keywords:	Climate Change, Cox Proportional Hazards Model, Explainable AI, Data Science, Climate Resilience, And Policy Support
DOI:	https://doi.org/10.5281/zenodo.18454388
Source:	Journal of Theoretical and Applied Information Technology 31^st January 2026 -- Vol. 104. No. 2-- 2026
Full Text

Title:	A HYBRID RESAMPLING–FEATURE SELECTION–ENSEMBLE FRAMEWORK FOR HIGH-FIDELITY CREDIT CARD FRAUD DETECTION
Author:	SRINIVASARAO DHARMIREDDI, VISHAL NAMIREDDY, SATHISH KUMAR DEEKONDA, M V V S SARMA, V. RAMA KRISHNA, RAVI KUMAR BOMMISETTI
Abstract:	Credit-card fraud detection requires high sensitivity to rare positive transactions while controlling false positives in large-scale, temporal transaction streams. Researcher design a reproducible ensemble framework that jointly optimizes resampling, hybrid feature selection, and model fusion to improve rare-event detection. Using a time-aware nested cross-validation protocol on the European credit-card benchmark, we treat resampling strategy and feature-selection as tunable components, apply SMOTE variants with cleaning, fuse filter / wrapper / embedded selection scores, and combine diverse base learners via stacking/weighted averaging. The proposed pipeline raises PR-AUC from 0.185 (best single baseline) to 0.322 and increases fraud recall from 0.56 to 0.78 (Wilcoxon p=0.012; McNemar p=0.0068) while remaining stable across seeds (PR-AUC CoV ≈ 2.8%) and robust to prevalence shifts and moderate input noise. Implications: These gains translate to materially fewer undetected frauds and manageable analyst workload, making the approach attractive for production deployment with appropriate monitoring and periodic retraining and model explainability checks.
Keywords:	Credit-Card Fraud; Class Imbalance; Resampling; Feature Selection; Ensemble Learning.
DOI:	https://doi.org/10.5281/zenodo.18454417
Source:	Journal of Theoretical and Applied Information Technology 31^st January 2026 -- Vol. 104. No. 2-- 2026
Full Text

Title:	TRANSFORMING MOROCCAN E-COMMERCE THROUGH GENERATIVE AI: EVIDENCE FROM AN EMPIRICAL STUDY
Author:	HAFSA LEMSIEH, ABDENNOUR CHATTAOU, AICHA ESSAOULA, YOUNESS JOUILIL, MUSTAPHA BASSOUR, JAMILA JOUALI
Abstract:	This research investigates the potential role of generative artificial intelligence (AI) within the e-commerce industry. Generative AI is employed in a variety of applications, such as automating content creation and personalizing user experiences, which significantly improve operational efficiency and customer engagement for e-commerce companies. For example, AI-driven product recommendations tailored to individual customer preferences have been shown to increase conversion rates, while AI-powered chatbots help you keep your customers satisfied and loyal with 24/7 customer service. However, the existing academic literature tends to focus on technological capabilities of AI in developed markets, leaving a considerable gap in understanding how these technologies are implemented and perceived in emerging economies such as Morocco. This paper seeks to address how e-commerce businesses can overcome specific barriers in order to effectively harness AI. Despite the obvious benefits, several challenges stand in the way of widespread adoption of generative AI in the e-commerce landscape. These include inadequate technology infrastructure, a shortage of qualified AI professionals, and significant ethical and regulatory concerns. Issues relating to data protection, privacy and transparency of AI algorithms are particularly pressing, requiring rigorous scrutiny and strong regulation frameworks to guarantee the ethical implementation of AI. The purpose of this study lies in the need to bridge the gap between worldwide progress in AI and national implementation strategies, in order to ensure that the digital transformation is adapted to the Moroccan context, both culturally and economically. For this study, a qualitative research was conducted through in-depth interviews with e-commerce professionals in Morocco. The analysis of the results was made by Maxqda Software. The results indicate that generative AI can boost efficiency, improve customer satisfaction and extend market reach. The study concludes that while challenges remain, the strategic implementation of generative AI, supported by appropriate policies and frameworks, can act as a catalyst for growth and innovation in e-commerce, positioning it competitively in the global marketplace. This research lays the foundations for further studies into the impact of AI on e-commerce, and offers practical recommendations for stakeholders wishing to leverage generative AI for business growth in Morocco.
Keywords:	Generative AI, E-Commerce, Empirical Study, Technology Acceptance.
DOI:	https://doi.org/10.5281/zenodo.18454427
Source:	Journal of Theoretical and Applied Information Technology 31^st January 2026 -- Vol. 104. No. 2-- 2026
Full Text

Title:	RESIDUAL ATTENTION-ENHANCED MULTI-SCALE DEEP LEARNING FRAMEWORK FOR ROBUST AND CONTINUITY-AWARE LUNG TUMOR SEGMENTATION FROM CT IMAGES
Author:	PALYAM MUTHYALA MADHURI, DR. VIJAY KUMAR DAMERA
Abstract:	The accurate segmentation of lung tumors in computed tomography (CT) scans is a critical step in computer-aided diagnosis and clinical planning. However, many existing segmentation pipelines remain sensitive to scale variability, boundary ambiguity, and inter-slice inconsistency, which can lead to fragmented masks and elevated false positives in challenging CT cases. However, complex anatomical variations, irregular lesion boundaries, and low tissue contrast often limit the performance of conventional deep learning models. This study introduces a Residual Attention-Enhanced Multi-Scale Deep Learning Framework (RAEMD-Net) designed to achieve robust and precise segmentation of lung tumors from CT images. The framework integrates multi-scale convolutional feature extraction, DenseNet-based residual connectivity, and a bidirectional LSTM module to capture both spatial and contextual dependencies within 3D image sequences. An adaptive attention mechanism dynamically adjusts the feature weights to enhance tumor boundary localization, while a hybrid loss function combining Dice and focal components improves gradient stability and class balance. Extensive experiments conducted on the LUNA16 and NSCLC-Radiomics datasets demonstrate that RAEMD-Net achieves a mean Dice coefficient of 0.96, Jaccard index of 0.94, and Hausdorff distance reduction of 12%, outperforming U-Net, SegNet, and Transformer-based baselines. Ablation analyses further indicate that residual attention contributes most strongly to false-positive suppression, while multi-scale fusion supports edge continuity across heterogeneous tumor sizes. The proposed model achieves computational efficiency through dynamic learning rate scheduling and parameter optimization, making it suitable for real-time deployment in medical imaging systems.
Keywords:	Deep Learning; Multi-Scale CNN; Residual Attention; LSTM Networks; Lung Tumor Segmentation; CT Imaging; Computer Vision; Hybrid Loss Optimization; Scientific Computing
DOI:	https://doi.org/10.5281/zenodo.18454439
Source:	Journal of Theoretical and Applied Information Technology 31^st January 2026 -- Vol. 104. No. 2-- 2026
Full Text

Title:	FAULT-TOLERANT CLUSTERING PROTOCOL FOR IOT WITH COLD AND HOT STANDBY REDUNDANCY
Author:	DR. A RADHA KRISHNA, DR. A LAKSHMI NARAYANA, KLV PRASAD, DR. INAKOTI RAMESH RAJA, G SANTHOSH KUMAR, NEHA BELWAL
Abstract:	IoT networks are becoming a more unsteady and heterogeneous setting in which frequent failure of devices, multi-modal noisy signals and uneven workloads diminish the stability of the cluster and the availability of the system. Although the current developments in the area of IoT analytics are rather fast, the currently used clustering protocols seldom incorporate adaptive redundancy, as well as do not resolve the problem of large-scale, error-prone deployment. The ability of this gap limit to be used in critical applications like smart cities, healthcare monitoring, and industrial IoT. The objective of this study was to create a robust and fault-tolerant clustering protocol, which is able to learn stable cluster-based structures in addition to dynamically maintaining cold and hot standby redundancy to ensure high availability with low overhead. A federated experimental model was adopted with publicly available IoT data (telemetry, network flows, acoustic faults, and ambient sensing), in which clustering was achieved with the help of Federated Graph-Contrastive Clustering (FGCC), and redundancy choices were made with the assistance of reinforcement learning. Devices were tested with injected failures, partitions, and variation of workloads to test the availability, MTTD, MTTR, clustering accuracy, and computational cost. The FT-CoH model was proposed to perform better, making ARI 0.78, availability 99.1, and the reduction of MTTR to 2.1 seconds, which is better in comparison to k-means, deep clustering, and federated baselines. These results show that adaptive cold happens to be much more efficient in terms of distributing fault tolerance in the IoT by means of multimodal federated clustering, which is additionally complemented by adaptive cold-hot standby management. In general, FT-CoH offers a reliable and scalable smart framework of IoT systems of the next generation, and the prospects it brings to autonomous monitoring, smart infrastructure, and critical IoT applications have been high.
Keywords:	Fault-Tolerant IoT, Federated Clustering, Hot–Cold Standby, Graph Neural Networks, Redundancy Optimization.
DOI:	https://doi.org/10.5281/zenodo.18454456
Source:	Journal of Theoretical and Applied Information Technology 31^st January 2026 -- Vol. 104. No. 2-- 2026
Full Text

Title:	AN EMPIRICAL STUDY ON THE IMPACT OF AI-BASED INSTRUCTIONAL PROGRAMS ON IMPROVING STORY WRITING SKILLS AMONG TENTH-GRADE STUDENTS
Author:	SAMEER ABDULSALAM ALSOUS, AWS I. ABUEID
Abstract:	The study aimed to reveal the effect of artificial intelligence programs on improving story writing among tenth-grade students. The study employed an experimental method, and the study sample consisted of one group of 23 female students who were randomly selected in a cluster sampling manner. The study used a pre-and post-test. The results showed a statistically significant difference (α = .05) for the students' scores on the story writing skills test before and after the application. Those who were exposed to the teaching strategy based on artificial intelligence programs in writing skills, combined with each skill, showed differences in favor of the post-application. The study made several recommendations, the most important of which was the use of artificial intelligence programs to enhance students' writing and storytelling skills.
Keywords:	Creative Writing, Artificial Intelligence, Magic School, Writing Skills, Education Technology, AI-Powered Learning
DOI:	https://doi.org/10.5281/zenodo.18454465
Source:	Journal of Theoretical and Applied Information Technology 31^st January 2026 -- Vol. 104. No. 2-- 2026
Full Text

Title:	INNOVATIVE PRACTICES OF PUBLIC ADMINISTRATION: APPROACHES TO PUBLIC SECTOR MANAGEMENT IN THE ERA OF DIGITALIZATION AND OPEN GOVERNMENT
Author:	OLEKSANDR SOMAK, IHOR DMYTRENKO, INNA NINYUK, INNA SURAY, OLEKSANDR NAKONECHNYI
Abstract:	The relevance of the study is due to the need for systematic analysis and implementation of the latest public administration practices that ensure the efficiency, transparency and adaptability of public institutions in the context of digitalization, globalization challenges and growing expectations of citizens. The research problem is the insufficient consideration of regional specifics and the effectiveness of renewable energy technologies implementation in the context of energy transformation. The purpose of the study is to systematize the latest public administration practices, identify the factors of their effectiveness, and develop an approach to innovation transfer, taking into account the challenges of digitalization and open government. The research results showed that modern public administration is formed at the intersection of technological innovations and humanistic values, where digitalization is not seen as an end in itself, but as a tool for enhancing transparency, inclusiveness, and trust in the state. A comparative analysis of international and national experience has shown that leading countries focus on creating integrated platform ecosystems and using artificial intelligence, while Ukraine emphasizes mobile services, paperless technologies, and active citizen engagement. Based on the clustering of EGDI dynamics in European countries (2010-2024), the author reveals the uneven development of digital governance and identifies four typical trajectories – from stable leaders to states that are in search of institutional prerequisites for digitalization. It is proved that the universal factors of success are political and institutional leadership, human and financial capacity, regulatory adaptability, cyber resilience, and civic participation. The importance of the humanistic dimension – the ability of innovative practices to reduce digital inequality, ensure social justice and maintain partnership between the state and citizens – is substantiated. Uneven development of digital public administration, limited inclusiveness, and insufficient institutional capacity to implement the latest management practices in the context of sustainable development and public trust have been identified. An effective model of public administration for the future will be based on the synergy of global technological trends and national contexts, combining innovation with a human-centered approach.
Keywords:	Digital Transformation, Innovative Practices, Administrative Management, EGDI, Cluster Analysis
DOI:	https://doi.org/10.5281/zenodo.18454478
Source:	Journal of Theoretical and Applied Information Technology 31^st January 2026 -- Vol. 104. No. 2-- 2026
Full Text

Title:	AUTOMATED RECOGNITION OF IRONY AND SARCASM IN ENGLISH FICTION USING NATURAL LANGUAGE PROCESSING TECHNIQUES
Author:	IRYNA BLYNOVA, NADIIA BRESLAVETS, YEVHENIIA ARTOMOVA, IVAN BAKHOV, VIKTORIIA MYKHAILENKO
Abstract:	Relevance of the research The relevance of the research is determined by the need for highly accurate automated detection of irony and sarcasm in literary discourse, which remains a challenge for modern Natural Language Processing (NLP) systems because of rhetorical complexity and semantic ambiguity. Aim of the research The aim of the research is to develop and formalize a multi-level methodology for building and validating a Hybrid NLP model for automated recognition of irony and sarcasm in English literary prose, taking into account rhetorical, stylistic, and contextual semantic characteristics. Research methods The research employed the following methods: comparative qualitative analysis (in two iterations), representative explication, comparative metric analysis, and decomposition analysis for optimization. Obtained results The comparative metric study found that the optimized Hybrid NLP model, which combines BERT4Irony, RuleML, Defeasible Logic, semantic pre-processing and meta-ensemble (XGBoost, MetaSVM), demonstrates the highest accuracy (Accuracy = 0.96, F1 = 0.945, AUC-ROC = 0.96) and adaptability to rhetorical complexity. The improvement of generative and discriminative metrics (BLEU = 0.91, ROUGE = 0.92, SPD = 0.88) confirms the model’s ability to accurately detect ironic inversions, stylistic deviations, and pragmatic implicits in literary discourse. Academic novelty of the research The academic novelty of the research is the developed Hybrid NLP model for automated irony and sarcasm detection, which combines BERT4Irony, ontological modelling (Defeasible Logic, RuleML), and meta-ensemble (XGBoost, MetaSVM). The model provides semantic accuracy and cognitive correspondence to the rhetorical structure of a literary text. Prospects for further research Further research may focus on the cross-linguistic adaptation of the model, taking into account linguistic and cultural differences in the conveyance of irony and sarcasm, in particular through the formalization of typical rhetorical patterns in different languages.
Keywords:	Irony, Sarcasm, Ontological Modelling, Hybrid NLP Model, Transformer Architecture, Semantic Framing, Rule-Based Formalism, Explainable AI, Discourse.
DOI:	https://doi.org/10.5281/zenodo.18454495
Source:	Journal of Theoretical and Applied Information Technology 31^st January 2026 -- Vol. 104. No. 2-- 2026
s

Title:	AN AI-ENABLED WEIGHTED PRIORITY SCHEDULING ALGORITHM FOR REAL-TIME TELEMEDICINE TRIAGE USING IOMT DATA
Author:	S. HEMALATHA, KIRAN MAYEE ADAVALA, N. MUTHUVAIRAVAN PILLAI, S. K. SATYANARAYANA , MANISHA MADDEL, G. KRISHNA MOHAN
Abstract:	One of the widely developing technologies is the Telemedicine which is accepted by all the categories of the disease patient. Providing the treatment on time will safeguard human life; to improve telemedicine consultation, several technological algorithms were added with the consultation assignment between the doctor and patient based on the Human Health Criticality Score. Even though this algorithm has produced better performance, there is still a need to improve the algorithm to produce more efficiency. Especially in the critical scenario like Emergency field, by understanding the patient's emergency condition and assigning the priority to the consultation can safeguard the lifetime of the patient. This article proposed a solution for the critical patient timely teleconsultation support. This research article proposed an algorithm named Enhanced Artificial Intelligent Priority-Based Telemedicine Algorithm (E-AIPTA), that algorithm automatically assigned the priority to the teleconsultation patient based on the real time physiological data. There are five primary body parameters are taken for assigning the priority named as heart rate (HR), oxygen saturation (SpO₂), blood pressure (BP), respiratory rate (RR), and body temperature (T) which are collected with the support of wearable sensor device attached with the patient. All these five parameters are assigned the medical parameter weight of low, medium, high, over-all Health Criticality Score (HCS) is computed to predict the emergency care patient. The proposed algorithm was evaluated using the simulated data sets and results were achieved over 93 % of accuracy, and average computation time is 0.84 seconds. The proposed work shows that it reduces latency and false alarm. This work can improve real time emergency management in critical illness people in teleconsultation.
Keywords:	Telemedicine, Teleconsultation, Heart Rate, Oxygen Saturation, Blood Pressure, Respiratory Rate, And Body Temperature, Health Criticality Score
DOI:	https://doi.org/10.5281/zenodo.18454514
Source:	Journal of Theoretical and Applied Information Technology 31^st January 2026 -- Vol. 104. No. 2-- 2026
Full Text

Title:	THE IMPACT OF ARTIFICIAL INTELLIGENCE ON THE TRANSFORMATION OF THE GLOBAL FINANCIAL MARKET MODELS
Author:	INNA KORSUN, ANDRIAN ULINICI, LILIA BUBLYK, ANDRII HUMENIUK, VITA ANDRIEIEVA
Abstract:	The relevance of the study is determined by the need for institutionalized rethinking of financial models under the influence of artificial intelligence (AI), which radically transform the architecture of decision-making, personalization, compliance, and sustainability mechanisms in the global FinTech environment. The aim of the study is to formalize and quantify the transformative impact of AI on financial market models by building a validation framework with stratified AI metrics. Research methodology: critical analysis of the implementation of AI technologies in FinTech, econometric modelling of process transformative AI technologies, structural decomposition analysis, Unified Modelling Language (UML) modelling of a stratified framework, econometric validation of an optimized AI architecture. An optimized FinTech AI reengineering framework with a cognitively stratified architecture was formed, which provides increased explainability (+19.3%), traceability (+22.1%), RLHF-adaptability adaptability (+16.7%), and resistance (+18.5%) while reducing latency (–14.2%) and achieving Compliance Automation Ratio (CAR) = 1.0. The academic novelty of the research is the developed AI architecture with the first-ever introduced transformation metrics (Automation Index, AI-Personalization Score, XAI Index, AITCI), which formalize the levels of algorithmization, personalization, explainability, and sustainability of FinTech models. The prospects for further research include a controlled testing with empirical stratification of each AI module of the FinTech architecture for a formal assessment of cognitive adaptability, regulatory traceability, and metric resilience in the real financial cycle.
Keywords:	AI-Enhanced Alpha; XAI Pipeline; RLHF Mechanisms; GNN Modules; Compliance Automation
DOI:	https://doi.org/10.5281/zenodo.18454530
Source:	Journal of Theoretical and Applied Information Technology 31^st January 2026 -- Vol. 104. No. 2-- 2026
Full Text

Title:	THE ROLE OF INFORMATION TECHNOLOGIES IN THE INTERNATIONAL LEGAL REGULATION OF PUBLIC SAFETY, LAW, AND ORDER
Author:	OLEKSANDR HOLOVKOV, MYKOLA TYSHLEK, IVO SVOBODA, IVAN KAYLO, ANDRII RYBALKIN
Abstract:	The fast pace at which information technology is being implemented in public administrations and security systems globally has dramatically changed the way public safety and law and order can be regulated internationally. Despite the extensive use of information technology by international organizations, there is very little scholarly work that provides empirical evidence of how digital technologies impact the efficiency, transparency, and legality of international security frameworks. This paper attempts to fill that void through a comparative empirical model based on a case study of the role of information technology in international legal regulation. This paper aims to assess how information technology impacts the performance of international legal frameworks that regulate public safety and law enforcement and also to identify the legal circumstances under which the digitalization process increases the regulatory effectiveness. To achieve these objectives, the authors have employed an interdisciplinary methodology using Legal Modelling of Standards, comparative doctrinal analyses and the Legal Impact Assessment Method in a controlled legal simulation environment (LexSim-Lab).Three international legal frameworks were analyzed as examples of international governance models - the UN Global Counter-Terrorism Strategy, the SIRIUS and SIENA platforms of Europol, and the OSCE cybercrime recommendations. The results show that the application of information technology clearly improves the time it takes to resolve conflicts, the level of procedural transparency and the level of adherence to norms; yet, the level of improvement varies across regimes. The results indicate that those regimes where the international legal framework was binding, institutionalized and/or had a high level of embeddedness, such as Europol, exhibited much better legal efficiency (up to 46.77%), greater transparency, fewer violations of norms and greater accountability than those regimes that were either politically coordinated or voluntary. Additionally, the results show that the digital tools used to support the various international legal frameworks acted as a "legal stress test" revealing structural weaknesses in international legal regulation that relate to accountability, jurisdiction and data sovereignty. The scientific originality of this study lies in the combination of legal and technological modeling approaches used to analyze the relationship between digitalization and measurable legal outcomes. Furthermore, the paper presents new knowledge by demonstrating that the effectiveness of international regulation of public safety depends less on the technological sophistication of the digital tool, than on the existence of enforceable legal architecture regulating the digital tool. Finally, the paper offers some practical advice for the development of international harmonized legal standards that can support secure, transparent and compliant with rights digital governance.
Keywords:	Information Technology, International Law, Public Safety, Law and Order, Legal Regulation, Cybersecurity Legislation, Data Protection, Security Policy
DOI:	https://doi.org/10.5281/zenodo.18454557
Source:	Journal of Theoretical and Applied Information Technology 31^st January 2026 -- Vol. 104. No. 2-- 2026
Full Text

Title:	REVISITING MOTIVATION IN DIGITAL LEARNING: THE INTERPLAY OF OERS INTERACTIVITY AND SELF-REGULATED LEARNING AMONG THAI PRE-SERVICE TEACHERS
Author:	PHENNAPA SUWANWONG, NURULLIZAM BINTI JAMIAT
Abstract:	Perceived motivation plays an essential role in digital learning, particularly for pre-service teachers who often struggle with self-paced environments, limited digital literacy, and insufficient self-regulated learning (SRL) skills. These factors can hinder their engagement and persistence in online courses. Interactive Open Educational Resources (I-OERs) have been proposed as a promising approach to enhance motivation because some of its features such as learner control, embedded questions, and feedback are presumed to support attention, confidence, and satisfaction. However, research exploring how interactivity interacts with different levels of learners’ self-regulated learning on their perceived motivation remains limited. Therefore, this study investigated main effects of Interactive Open Educational Resources (I-OERs) and Non-Interactive Open Educational Resources (N-I-OERs) and self-regulated learning as well as their interaction effect on perceived motivation among 117 pre-service teachers. Using a quasi-experimental design with a three by two factorial design, each student as determined by self-regulated learning questionnaire was categorized into high, moderate, and low groups and purposively assigned to study with either an I-OERs or N-IOERs course materials available on a MOOC platform. Then, perceived motivation was measured by using the Instructional Materials Motivation Scale (IMMS), and data was analyzed with two-way ANOVA and descriptive statistics including mean and standard deviation. The key findings showed that significant main effects of different learning materials and varying degrees in self-regulated learning on perceived motivation among participants were evidently observed. The students who used interactive materials and possessed higher self-regulated learning exhibited greater motivation. Additionally, an interaction effect between types of learning materials and regulated learning control significantly affected the rise in learning motivation. The study highlights the importance of incorporating interactive and motivational features into open educational resources while offering adequate support for learners with lower self-regulation.
Keywords:	Open Educational Resources; Perceived Motivation; Self-Regulated Learning; H5P; Multimedia Learning; Pre-Service Teachers
DOI:	https://doi.org/10.5281/zenodo.18454566
Source:	Journal of Theoretical and Applied Information Technology 31^st January 2026 -- Vol. 104. No. 2-- 2026
Full Text

Title:	CAUSALGRAPH-EMOTIONNET PERSONALITY-CONSCIOUS CAUSAL GRAPH TRANSFORMER WITH NEURAL-ODE TEMPORAL MODELLING TO EXPLAINABLE MULTIMODAL EMOTION RECOGNITION
Author:	T L DEEPIKA ROY, NULAKA SRINIVASU
Abstract:	Multimodal emotion recognition Multimodal physiological and behavioral emotion recognition is of critical importance in affective computing, human-computer interaction (HCI), and mental-health analytics. However, current deep learning models generally do not take modalities into account (disregarding their causal and temporal inter-dependencies) and ignore personality-based variability that is essential to realistic affect modelling. To address these shortcomings, the given paper proposes CausalGraph-EmotionNet, a personality-conscious causal graph transformer, which combines Neural-ODE-based temporal evolution with causal attention-assisted multimodal fusion. The AFFECT data of each modality (EEG, electrodermal activity, facial activity, eye gaze, pupil dilation, and cursor movement) is modeled as a dynamic causal graph the time-varying connectivity of which reflects time-varying functional and directional interactions. The merged embeddings are optimized by personality-conditioned causal attention systems, which allows making individualized and interpretable inferences about emotions. Large-scale experiments on the AFFET dataset indicate that CausalGraph-EmotionNet has 84.6% accuracy and 80.8% macro-F1, outperforming CNN, RNN, GCN, Transformer and PhysioGraph-Transformer. The model significantly enhances the identification of more complex affective conditions like fear and disgust, it is resilient to a 40% loss in modality and has interpretable causal maps that bridge personality dimensions and modality salience. The findings make CausalGraph-EmotionNet a state-of-the-art, explainable and causally motivated architecture of multimodal emotion recognition - a unification of data-driven learning with psychologically relevant causal inferences.
Keywords:	Affective Computing, High-Level Resources, Emotion Recognition, Causal Graph Transformer, Neural ODE, Personality-Sensitive Fusion, Multimodal Learning, Explainable AI Physiological Signals AFFEC Dataset.
DOI:	https://doi.org/10.5281/zenodo.18454574
Source:	Journal of Theoretical and Applied Information Technology 31^st January 2026 -- Vol. 104. No. 2-- 2026
Full Text

Title:	URBAN TRAFFIC GOVERNANCE USING DU-NET: A DUAL-STREAM DEEP LEARNING FRAMEWORK FOR REAL-TIME ROAD ANOMALY DETECTION AND INTELLIGENT VEHICLE ANALYTICS
Author:	VIJAY BHASKAR S, Dr. L. V. V. GOPALA RAO, Dr. A. GOPALA KRISHNA
Abstract:	The rapid rise of urbanization and the connected mobility demonstrates the serious limitations facing traditional traffic management systems. Solutions previously focused on vehicle flow. Others relied on manual inspection for road maintenance. Both these were not comprehensive. They also caused delays in answering urban transport issues. In this paper, we will introduce DU-Net (Dual-Stream Urban Network), an intelligent deep-learning framework for real-time detection of road anomalies and vehicle analytics in a smart city. The design utilizes high-definition cathode ray tube roadside camera visual sensing and IOT sensor data from embedded infrastructure for multi-task learning to detect potholes and classify vehicles and dynamically estimate traffic density. A convolutional pipeline with two streams can capture spatial dependencies and subsequent temporal dependencies. A probabilistic fusion model views the hypotheses for sensor signals and video signals as being generated from a probabilistic mixture model. It can align the video-based hypotheses with the sensor-based hypotheses to achieve consistent decisions. The system uses a smart technique that gives an appropriate weight to each task in order to enhance robustness over weather and lighting and traffic conditions. Experimental analysis using urban road datasets shows DU-Net to outperform state-of-the-art detectors like YOLOv8 and Faster R-CNN in terms of evaluation metrics, inference speed, and computational cost. Moreover, its prediction analytics component facilitates scheduling of maintenance and traffic congestion prediction. The suggested DU-Net framework provides an extensible foundation for smart transportation systems in the future settings through data-driven, adaptive and climate-resilient urban traffic governance.
Keywords:	Deep Learning, Dual-Stream Architecture, Intelligent Transportation Systems, Iot Data Fusion, Multi-Task Learning, Pothole Detection, Smart Cities, Traffic Analytics, Urban Governance, Vehicle Classification.
DOI:	https://doi.org/10.5281/zenodo.18454589
Source:	Journal of Theoretical and Applied Information Technology 31^st January 2026 -- Vol. 104. No. 2-- 2026
Full Text

Title:	MACHINE LEARNING FRAMEWORK FOR DYNAMIC PRICING: INTEGRATING LSTM FORECASTING AND REINFORCEMENT LEARNING OPTIMIZATION
Author:	MEHULKUMAR H KANTARIA, DR. MANJU SHARMA, DR. B SIVA LAKSHMI, S. KRISHNAKUMARI, MOLIGI SANGEETHA, DR. P. VENKATESWARA RAO
Abstract:	This manuscript addresses short-horizon retail pricing under misaligned mixed-frequency IoT signals (POS, footfall, energy) and the forecast–control representational mismatch. Dynamic pricing in retail requires accurate short-horizon demand estimation and constraint-aware policy optimization to reconcile revenue maximization with service levels. Existing pipelines frequently decouple forecasting and control, thereby incurring responsiveness and suboptimality under mixed-frequency telemetry regimes. This study aimed to develop and evaluate a hybrid pipeline that integrates MIDAS-style mixed-frequency covariate alignment with Long Short-Term Memory (LSTM) forecasting and an advantage actor–critic controller to improve profit, conversion and service metrics in retail/e-commerce settings. A two-stage protocol was executed: Mixed data sampling (MIDAS) logistic-weight aggregation converted IoT/edge streams into decision-cadence features. An LSTM encoder produced one-step forecasts that were (a) pretrained (b) optionally fine-tuned jointly with an A2C agent. Reward shaping incorporated discounted profit, inventory penalties and smoothness/fairness regularizers. Evaluation used synthetic nonstationary simulators and a composite POS+ footfall+ energy dataset with temporal train/validation/test splits. The joint MIDAS-LSTM + A2C pipeline yielded statistically significant profit uplifts (≈13–18% vs strong baselines), lower forecast MSE, reduced stockouts and smoother price trajectories. Ablation (no MIDAS, frozen LSTM, no smoothness) validated the contribution of each module. Integrating mixed-frequency telemetry into an end-to-end forecast-to-policy loop materially improves economic outcomes and operational robustness. Results indicate statistically significant profit uplifts, improved forecast accuracy, and smoother pricing, although elasticity identification and latency constraints remain practical limitations. Further work is required on causal elasticity identification and latency-constrained deployment.
Keywords:	Dynamic Pricing, Reinforcement Learning, Long Short- Term Memory, Mixed Data Sampling, Retail Analytics, Iot Mixed Frequency Data
DOI:	https://doi.org/10.5281/zenodo.18454609
Source:	Journal of Theoretical and Applied Information Technology 31^st January 2026 -- Vol. 104. No. 2-- 2026
Full Text

Title:	DNA-MAPPED OPTICAL CRYPTOGRAPHY FOR ROBUST AND EFFICIENT IOT SECURITY
Author:	VISHAL NAMIREDDY, DR. TAHMEENA FATIMA, DR. UTTARA GOGATE, MS.LEILA BENNJIMA, R S S RAJU BATTULA, G SANTHOSH KUMAR
Abstract:	DNA-based optical cryptography for Internet-of-Things (IoT) devices offer a path to secure low-power telemetry with minimal per-packet computation. Existing solutions rely on heavy symmetric ciphers that burden constrained microcontrollers (MCUs) or on bio-inspired encoders that lack scalable optical integration and rigorous real-time traces. This paper proposes DO-OPT (DNA-based Optical Cryptography), a hybrid framework that combined thermodynamically constrained DNA-code mappings, an optical phase-noise derived TRNG (true-random number generator), and a lightweight, post-quantum-capable permutation to enable authenticated, low-latency transmission for IoT telemetry. DO-OPT performed hardware-friendly DNA LUT (lookup table) mappings on the MCU, derived high-entropy keys from an optical TRNG passed through a KDF (key derivation function), and exploited photonic parallelization in an SLM-style optical encoder to reduce wall-clock per-packet processing. The method preserved high ciphertext entropy and increased Key Unpredictability Index (KUI) while lowering energy-per-byte. DO-OPT was evaluated on Gotham Dataset 2025 (78 heterogeneous IoT devices, 1.2M packets) and on a 1,000-node emulated deployment. Metrics included latency, energy-per-byte, KUI, NIST STS (Statistical Test Suite) outcomes, and robustness to ciphertext-only and replay attacks. DO-OPT reduced median per-packet latency by 37.5% and energy-per-byte by 29.2% relative to AES-128, increased entropy per byte from 7.10 to 7.98 bits versus the best prior DNA-inspired scheme, and passed full NIST STS validation. These results indicate DO-OPT provides a practical, energy-efficient security layer suitable for real-time IoT telemetry using DNA–optical hybrids.
Keywords:	DNA-Based Optical Cryptography, DNA Cryptography, Optical Computing, Iot Security, TRNG, KUI, Telemetry.
DOI:	https://doi.org/10.5281/zenodo.18454623
Source:	Journal of Theoretical and Applied Information Technology 31^st January 2026 -- Vol. 104. No. 2-- 2026
Full Text

Title:	AHDNAM: AN OPTIMISED HYBRID DEEP LEARNING AND SWARM INTELLIGENCE FRAMEWORK FOR BREAST ULTRASOUND DIAGNOSIS
Author:	DR. M. VAMSIKRISHNA, G RESHMA, NARENDRA B MUSTARE, B V SUBBA RAO, DRAKSHAYANI SRIRAMSETTI, ELANGOVAN MUNIYANDY
Abstract:	Breast tumor classification from ultrasound is challenged by low contrast and speckle noise, limiting reliable automated triage. A reproducible, high-recall classifier that is robust to device and acquisition variability remains a pressing gap for clinical workflows. This study aimed to evaluate an adapted BCDNet incorporating a dual-branch hybrid classifier and an adaptive constrained hyperparameter search to improve malignant lesion detection and overall balanced performance on public ultrasound datasets. An experimental study used ImageNet pretrained VGG16 as a backbone with two parallel classifier branches: a dilated atrous spatial pyramid pooling branch and a channel-attention branch. Augmentation (mixup, cutmix), label smoothing, focal loss, and stochastic weight averaging were applied. Patient-level splits from the Kaggle breast ultrasound dataset and BUSI were used for training, validation, and held-out testing; hyperparameters were tuned with a 50-evaluation RPAOSM-ESO-inspired search and top models ensembled. On held-out tests, the hybrid pipeline attained an accuracy 94.5% (Kaggle) and 93.2% (BUSI), a sensitivity ≈ 94.6% and 93.0%, an MCC of 0.881 and 0.860, false negative rates 5.43% and 6.80%, and an AUC ≈ of 0.966 and 0.959, respectively. Bootstrap CIs indicated statistically meaningful gains in MCC and FNR versus single-branch baselines. In spite of the fact that, before the proposed pipeline can be utilized in clinical practice, more comprehensive annotations and broader multi-center validation have to be taken into consideration, it proves that it can be used as a research tool that results in a smaller number of missed malignancies without compromising the specificity.
Keywords:	Breast Ultrasound, Transfer Learning, VGG16, ASPP, Attention Hybrid Classifier, Hyperparameter Optimization, Explainability
DOI:	https://doi.org/10.5281/zenodo.18454647
Source:	Journal of Theoretical and Applied Information Technology 31^st January 2026 -- Vol. 104. No. 2-- 2026
Full Text

Title:	CONTEXT-AWARE IMAGE RETRIEVAL: ENHANCING SEARCH PRECISION IN LARGE- SCALE IMAGE DATABASES USING BLIP AND AUTOMATED CAPTIONING
Author:	JAHNAVI SOMAVARAPU, RAVI KANTH MOTUPALLI, ANJANEYULU NELLURU, VENKATESWARA RAO KOTA, SUDHAKAR YADAV NALADESI, TEJASWI POTLURI
Abstract:	The growing number of digital images has rendered efficient retrieval a serious issue, since conventional approaches based on low-level features like color and texture are not able to capture semantic content. Users often find it difficult to retrieve appropriate images because there is no context-aware search mechanism, particularly when precise metadata or visual information is not known. Current retrieval systems are unable to fill the semantic gap between text-based queries and image content, producing inaccurate or incomplete results. In order to solve this, we introduce a context-aware image retrieval sys- tem that combines deep learning and natural language processing (NLP) approaches. The system utilizes BLIP model for automatic image captioning, constructing a meaningful textual description of images. User requests are processed based on Sentence-BERT (SBERT) embeddings, coupled with TF-IDF and Levenshtein distance-based string matching to enable accurate retrieval. An intuitive user-friendly search is offered through a Gradio-based interface. Through this work, it is illustrated that visual feature extraction combined with NLP captioning and query processing boosts the accuracy of image search substantially. The system suggested here not only enhances the precision in retrieval but also maintains scalability and adaptability for big unstructured data. Future efforts will be dedicated to achieving further computational efficiency, including multimodal retrieval, and improving user personalization to improve the system responsiveness to various user needs.
Keywords:	Image Retrieval, Semantic Search, Deep Learning, Sentence- BERT, Query Encoding, Context-Aware Search, Gradio Interface.
DOI:	https://doi.org/10.5281/zenodo.18454673
Source:	Journal of Theoretical and Applied Information Technology 31^st January 2026 -- Vol. 104. No. 2-- 2026
Full Text

Title:	DEVELOPMENT OF A REAL-TIME AND RETROSPECTIVE INTELLIGENT OBJECT DETECTION SYSTEM FROM CCTV FOR SECURITY ENHANCEMENT IN LOCAL GOVERNMENT ORGANIZATIONS
Author:	CHANVIT PHROMPHANCHAI, SUWUT TUMTHONG
Abstract:	This research develops an AI-RTRO system to address the limitations of traditional CCTV in Bang Yai Municipality's Smart Safety Infrastructure. The system improves real-time situational awareness and enhances retrospective evidence retrieval for local government. Managing large-scale CCTV data and making rapid decisions are key operational challenges. The system integrates the AI Agent Core framework and IPOF model for object detection, tracking, and retrieval from both real-time and archived video. Deep learning algorithms (YOLO, DeepSORT/ByteTrack, ResNet, Faiss) are used, with Grad-CAM providing transparency. Performance is evaluated in real-world conditions with mAP, IDF1, and latency. Results show accurate detection, stable tracking, and low-latency operation. Metadata management and feature vector indexing improve the performance of retrospective retrieval. The study confirms that integrating AI Agent Core and IPOF improves system performance, retrieval efficiency, decision transparency, and continuous learning. The framework was rated highly by 15 experts, with a mean of 4.74 (S.D. = 0.55), indicating readiness for practical use in local government surveillance.
Keywords:	Smart City, Object Detection, Time-Indexed Retrieval, CCTV Analytics, AI Agent Core
DOI:	https://doi.org/10.5281/zenodo.18454735
Source:	Journal of Theoretical and Applied Information Technology 31^st January 2026 -- Vol. 104. No. 2-- 2026
Full Text

Title:	ALGORITHMIC DETECTION OF ANOMALIES IN PRODUCTION LOGS: DEV-QA STRATEGIES LEVERAGING ARTIFICIAL INTELLIGENCE AND TRADITIONAL METHODS
Author:	OLEH SYPIAHIN, ANNA DEVIATKO, VOLODYMYR LOPUKHOVYCH, TYHRAN OVSEPIAN, OLEKSANDR SHVAIKIN
Abstract:	Relevance: The relevance of the research is determined by the need to increase the efficiency of Dev-QA pipelines by integrating automated testing, anomaly detection, and artificial intelligence (AI)-oriented algorithms to ensure stable software quality. Aim: The aim of the research is to formalize, model, and metrically verify an optimized Hybrid Pipeline to increase the automation, accuracy, and speed of bug reporting and anomaly detection in Dev-QA. Methods: Methodological basis of the study: cognitive-functional analysis, metric modelling, Unified Modelling Language (UML)-based modelling Rule-Based Pattern Matching (AI-less), UML-based modelling Hybrid Pipelines (AI-driven), decomposition and optimization analysis, UML modelling optimized Hybrid Pipelines, Iterative metric-based modelling. Results: Optimized Dev-QA Hybrid Pipeline with integration of rule-based pre-filtering, ML post-classification, Human-in-the-loop (HITL) augmentation and auto-learning increased anomaly detection efficiency by 17.4%, reduced MTTR by 21.8%, reduced communication overhead by 15.6%. Achieved Precision = 0.94, Recall = 0.92, F1-score = 0.93, Coverage = 96%, fault tolerance = 99.1% uptime, reduction of repeated defects — 18.2%. Academic novelty: The academic novelty of the research is the formalization of the Dev-QA Hybrid Pipeline with cognitive stratification of anomaly detection and bug reporting, integration of rule-based and machine learning (ML) modules with HITL augmentation and auto-learning for metrically validated end-to-end automation. Prospects for further research: Further research prospects include experimental implementation of the optimized Dev-QA Hybrid Pipeline in a controlled Continuous Integration/Continuous Delivery or Deployment (CI/CD) environment with subsequent analysis of its technical and procedural efficiency to justify scaled implementation and regulatory institutionalization.
Keywords:	Test Automation, Anomaly Detection, Rule-Based Algorithms, Machine Learning, Dev-QA Pipelines, Bug Reporting, Software Quality Assurance
DOI:	https://doi.org/10.5281/zenodo.18454753
Source:	Journal of Theoretical and Applied Information Technology 31^st January 2026 -- Vol. 104. No. 2-- 2026
Full Text

Title:	DETECTION OF MEANINGFUL COMMUNITIES IN SOCIAL LEARNING NETWORKS
Author:	HICHAM SADIKI, RAJAE ZRIAA,AYOUB ENNASSIRI, MAHMOUD LHAM, SAID AMALI
Abstract:	Community detection and analysis in social learning networks is a critical challenge for understanding collaborative dynamics. These networks primarily rely on two distinct structures: pedagogical interests and social interactions. A promising approach to uncover meaningful learning communities is to combine these two structures. In this context, we propose a hybrid and adaptive two-stage approach that integrates both structural and contextual information to efficiently detect and update communities in dynamic educational social networks. This approach proceeds in two stages. First, we introduce a static algorithm called Learning-Enhanced Method for Community Detection (LEMCD). This hybrid algorithm leverages statistical and semantic measures to analyze both the pedagogical content associated with learners and their social interactions. Second, we present Adaptive LEMCD (LEMACD), an adaptive algorithm for detecting and updating community structures in dynamic social networks. A comparative study was conducted using real academic networks to evaluate the performance of LEMACD against other content- and behavior-based community detection algorithms. The results demonstrate that our approach is capable of identifying coherent communities in social networks. Moreover, it adapts effectively to changes in learning networks while reducing response time.
Keywords:	Dynamic Social Networks, Modularity, Content Information, Topic Detection, Meaningful Community Detection, Social Learning Networks.
DOI:	https://doi.org/10.5281/zenodo.18454778
Source:	Journal of Theoretical and Applied Information Technology 31^st January 2026 -- Vol. 104. No. 2-- 2026
Full Text

Title:	DEEP LEARNING APPROACHES FOR CARDIAC DISEASE DIAGNOSIS USING MULTI-MODAL INTEGRATION
Author:	SUNEETHA DAVULURI , SIVANEASAN BALA KRISHNAN, PRASUN CHAKRABARTI, VENUGOPAL BOPPANA,SRI HARI NALLAMALA
Abstract:	This research presents a novel quantum-hybrid machine learning framework for heart disease classification that integrates Quantum Neural Networks with variational circuits and Quantum Support Vector Machines featuring custom quantum kernels alongside classical ensemble methods. The methodology employs advanced feature engineering, multi-strategy missing value imputation, and combined feature selection techniques on an enhanced dataset of 2,000 cardiovascular cases with 17 engineered features. The quantum components utilize 8-10 qubits across 5-6 variational layers with quantum interference effects for enhanced pattern recognition capabilities. Through weighted soft voting ensemble strategy and dynamic threshold optimization, the quantum-hybrid system achieved exceptional performance metrics including 96.2% accuracy, 95.8% F1-score, 94.7% precision, and 96.9% recall, significantly exceeding the 95% accuracy target. The quantum models demonstrated measurable advantages over classical approaches, with the Quantum Neural Network reaching 92.4% accuracy and Quantum SVM achieving 90.8% accuracy independently. The hybrid ensemble showed 4.3 percentage points improvement over the best individual classical model and 2.7 percentage points over standalone quantum implementations. This quantum-hybrid approach establishes superior diagnostic capabilities for cardiovascular disease prediction, particularly in minimizing false negatives critical for patient safety, positioning quantum-enhanced machine learning as a viable solution for next-generation medical diagnostic systems with clinically relevant accuracy improvements for healthcare applications.
Keywords:	Quantum Machine Learning, Cardiovascular Disease Classification, Hybrid Ensemble Methods, Variational Quantum Circuits, Medical Diagnostic Accuracy, Quantum-Classical Integration
DOI:	https://doi.org/10.5281/zenodo.18454790
Source:	Journal of Theoretical and Applied Information Technology 31^st January 2026 -- Vol. 104. No. 2-- 2026
Full Text