		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
December 2025 | Vol. 103 No.24

Title:	OPTIMISED CUCKOO-INSPIRED DISCRETE SYMBIOTIC ORGANISMS SEARCH STRATEGY FOR TASK SCHEDULING IN A CLOUD COMPUTING ENVIRONMENT
Author:	SULEIMAN SAAD, ABDULLAH MUHAMMED, MOHAMMED ABDULLAHI , AZIZOL ABDULLAH , FAHRUL HAKIM AYOB
Abstract:	The cloud computing paradigm is rapidly expanding as there is a notable shift away from other distributed computing approaches and traditional IT infrastructures. As a result, optimised task scheduling techniques have become essential for managing the growing complexity of cloud environments. In cloud computing, numerous tasks must be allocated to a limited number of diverse virtual machines, aiming to reduce the imbalance between local and global search spaces while optimising system utilisation. Task scheduling is an NP-complete problem, meaning there is no exact solution, and we can only achieve near-optimal results, especially when dealing with large-scale tasks in cloud computing. This paper introduces an optimised strategy, the Cuckoo-based Discrete Symbiotic Organisms Search (C-DSOS), for optimal task scheduling in cloud environments. The method is based on the Standard Symbiotic Organism Search (SOS), a nature-inspired metaheuristic optimisation algorithm used for numerical optimisation problems. SOS mimics the symbiotic relationships seen in ecosystems, such as mutualism, commensalism, and parasitism. The proposed technique was evaluated using the CloudSim toolkit simulator, and experimental results showed that C-DSOS outperforms the benchmarked Simulated Annealing Symbiotic Organism Search (SASOS) algorithm, commonly used in task scheduling. C-DSOS demonstrated a better convergence rate, especially in larger search spaces, making it well-suited for cloud task scheduling. Additionally, a t-test analysis revealed that C-DSOS is statistically significant compared to SASOS, particularly in scenarios involving larger search spaces.
Keywords:	Cloud computing, Scheduling, Metaheuristic, C-DSOS, Optimisation
Source:	Journal of Theoretical and Applied Information Technology 31^st December 2025 -- Vol. 103. No. 24-- 2025
Full Text

Title:	EFFICIENT DEEP LEARNING WITH COMPRESSED MUZZLE PRINTS FOR SCALABLE BUFFALO IDENTIFICATION
Author:	FATIMAH BINTI KHALID, RANA RANJEET SINGH, NURUL AIN SHAFFERI, TIMUR R AHLAWAT, M S SANKANUR, ANURADHA AGRAWAL, PRERNA GHORPADEG, SITI KHADIJAH ALI, NURUL AMELINA NASHARUDDIN, MAS RINA MUSTAFFA
Abstract:	Traditional livestock identification methods, such as ear tagging and branding, are constrained by scalability, security, and welfare issues. Biometric identification, specifically muzzle pattern recognition, is a non-invasive and tamperproof alternative. Deploying deep learning models for muzzle pattern recognition is constrained by the need for substantial dataset sizes, storage requirements and longer training time. This study introduces an approach that integrates extreme data compression with lightweight deep learning models to enable efficient buffalo identification. Unlike earlier studies that mainly focused on classification accuracy, our study provides a thorough examination of the effects of compression on model performance and storage capabilities. A dataset of 49GB was compressed using three techniques: Principal Component Analysis (PCA), K-Means clustering, and JPEG compression, which successfully reduced the storage size while retaining essential features. PCA demonstrated the most effective compression, achieving the highest compression ratio (2585.69×) and shrinking the dataset by 99.86% to 68.6MB while sustaining high accuracy. Three lightweight deep learning models, MobileNetV2, RegNetY_400MF, and SqueezeNet 1.1, were trained and evaluated on both compressed and uncompressed datasets. The results demonstrate that MobileNetV2 and RegNetY_400MF maintained over 99% accuracy after compression, which suggests that extreme data reduction does not substantially affect recognition performance. This study presents one of the initial practical applications of a compressed biometric dataset for livestock identification. The system was deployed as a Flask-based web application, MuzzleID, enabling practical usage in livestock management. Our findings suggest that reducing the size of datasets significantly enables the widespread implementation of biometric systems in regions with limited resources while ensuring precise identification at a high standard.
Keywords:	Muzzle Recognition, Biometric Identification, Data Compression, Lightweight Deep Learning, Buffalo Identification, PCA, Mobilenetv2, Livestock Management
Source:	Journal of Theoretical and Applied Information Technology 31^st December 2025 -- Vol. 103. No. 24-- 2025
Full Text

Title:	DESIGN AND DEVELOPMENT OF HYBRID ALGORITHMS FOR QUANTUM - HIGH PERFORMANCE COMPUTING ARCHITECTURES
Author:	Mrs. .B.V. PRASANNA LATHA , Dr. MUKTEVI SRIVENKATESH
Abstract:	As quantum computing steadily advances within the Noisy Intermediate-Scale Quantum (NISQ) era, the integration of Quantum Processing Units (QPUs) with classical High-Performance Computing (HPC) systems emerges as a compelling strategy to address computationally intensive and complex problems across various domains. However, the limitations of current quantum hardware — such as limited qubit counts, decoherence, and gate errors — necessitate hybrid approaches that combine quantum and classical computational resources in an intelligent and efficient manner. This paper presents a novel approach to the design and development of hybrid algorithms that enable dynamic partitioning of computational workloads across heterogeneous computing subsystems, including Central Processing Units (CPUs), Graphics Processing Units (GPUs), Artificial Neural Network (ANN) accelerators, and QPUs. The central goal is to harness the complementary strengths of these architectures: CPUs for control flow and data orchestration, GPUs for massively parallel computations, ANN accelerators for low-latency neural processing, and QPUs for solving specific subproblems such as combinatorial optimization, quantum simulations, and enhanced kernel computations in machine learning. We propose a generalized framework for dynamic task partitioning that analyzes workload characteristics — such as data dependencies, computational complexity, and hardware affinity — to automatically distribute tasks to the most appropriate hardware components. Furthermore, we introduce a hybrid scheduler capable of adapting task allocations at runtime based on system load and quantum execution latency. To validate our approach, we present several case studies across diverse application domains: combinatorial optimization (Max-Cut problem), quantum chemistry simulation (Variational Quantum Eigensolver), and quantum-enhanced machine learning (Quantum Kernel Support Vector Machines). Empirical results demonstrate that our hybrid architecture delivers significant speedup and energy efficiency gains compared to both conventional classical HPC pipelines and pure quantum or classical approaches. By providing a systematic framework for hybrid algorithm design, this work contributes toward making practical hybrid Quantum-HPC systems a viable paradigm for solving real-world problems as quantum hardware continues to evolve.
Keywords:	High Performance Computer(HPC), Quantum Approximate Optimization Algorithm (QAOA), Quantum Machine Learning (QML),Heterogeneous Computing Architectures(HCA), Quantum-Classical Workload Scheduling(QCWS)
Source:	Journal of Theoretical and Applied Information Technology 31^st December 2025 -- Vol. 103. No. 24-- 2025
Full Text

Title:	TOWARDS SMART AGRICULTURE: AN OPTIMAL ENSEMBLE DEEP LEARNING SCHEME FOR BANANA LEAF NUTRIENT DEFICIENCY IDENTIFICATION
Author:	REKHA.V , DR. UMA SHANKARI SRINIVASAN
Abstract:	Crop’s Nutrition is highly essential for health conditions at the time of its growth phase and yield. The lack or shortage of needed nutrients is regraded one such crucial factor that impacts overall yield of crop. A novel approach for automated recognition and classification of banana leaf nutrition deficiency is presented using Ensemble Deep Learning approach. Initially, the input images are pre-processed, filtered and augmented. It is then segmented using Improved TGV-FCM (Total generalized variation-Fuzzy C-Means) scheme. From the segmented image, the features are extracted using Residual GCN (graph convolution network) followed by selection of suitable optimal features through Enriched Glow swarm (EGSO) optimization approach. Ranking based Ensemble MobileNet v2 and LeNet classifier model is used for classification for diagnosing nutrition deficiency in banana leaf. The validation of proposed model is estimated and are compared with traditional models in terms of various metrics to exemplify the efficacy of presented model.
Keywords:	Banana crop, Nutrition Deficiency Detection, Deep Learning, Residual Graph Convolution Network, Enriched Glow Swarm Optimization, Ranking based Ensemble MobileNet v2 and LeNet classifier
Source:	Journal of Theoretical and Applied Information Technology 31^st December 2025 -- Vol. 103. No. 24-- 2025
Full Text

Title:	ORGANIZATIONAL AND ANALYTICAL SUPPORT FOR CONDUCTING EXTERNAL AUDITS OF DATA PROTECTION SYSTEMS IN INSTITUTIONS
Author:	MYKOLA MASESOV, OLEG DIEGTIAR, DMYTRO MINOCHKIN2, ANDRII KHAPSALIS, DMYTRO NOVYTSKYI, SERGII KRAMARENKO
Abstract:	The main goal of information security is to protect information and related infrastructure from damage that may arise as a result of loss, alteration, or unauthorized distortion of information, whether due to criminal attempts or accidental actions. In such cases, the goal is to minimize the damage that may be caused by such circumstances. From time to time, it is also advisable to conduct an independent assessment of the information security in place at the organization. Such audits work at the level of actual protection and response to risks that are constantly changing. An audit is, in essence, an indispensable source of further recommendations for the design and modernization of protection systems to ensure compliance with standards. This article outlines the main stages of the methodology for developing an analysis of the effectiveness of information security measures in organizations, although some of the proposed indicators require serious refinement, which opens up new opportunities for further scientific research.
Keywords:	Independent Information Security Audit, Information Security Auditor, Information Security Assessment, Cybersecurity Assessment
Source:	Journal of Theoretical and Applied Information Technology 31^st December 2025 -- Vol. 103. No. 24-- 2025
Full Text

Title:	DEVELOPMENT OF INTELLIGENT ADVISORY SYSTEM WITH COGNITIVE TECHNOLOGY
Author:	PHANINTORN SUAPRAE, PRACHYANUN NILSOOK, PANITA WANNAPIROON, VITSANU NITTAYATHAMMAKUL
Abstract:	Existing research on student retention mainly focuses on risk prediction, with few studies implementing advisory processes that translate predictions into timely, personalized interventions. This study develops and evaluates the intelligent advisory system with cognitive technology (referred to as the IAS-CT system) to improve student retention in higher education. The persisting gap in the literature is that most retention studies stop at risk prediction and rarely operationalize a closed-loop advisory workflow that converts predictions into timely, personalized interventions. Using de-identified institutional records of 2,973 undergraduates from academic years 2019–2022 with 25 academic and socio-demographic features, we trained and compared Decision Trees, Logistic Regression, Random Forest, K-Nearest Neighbors, and Naive Bayes. Preprocessing comprised imputation, normalization, and categorical encoding/selection; evaluation used a stratified split and standard metrics (accuracy, precision, recall, and F1) with confusion matrices. Correlation analysis indicated that GPA (r = 0.55), absenteeism (r = 0.48), father’s income (r = 0.45), year of study (r = 0.38), and field of study (r = 0.20) were the most associated factors with retention. Decision Trees achieved the best predictive performance (accuracy = 98.90%), exceeding Logistic Regression (97.40%), Random Forest (86.10%), K-Nearest Neighbors (85.90%), and Naive Bayes (85.80%). The selected model was integrated into an advisory architecture that issues early-warning alerts, generates personalized study recommendations, and supports advisor–student communication. An expert panel rated the system’s suitability at an overall high level. Consequently, the system operationalizes prediction into intervention, providing actionable retention support with practical implications for data governance and institutional scaling.
Keywords:	Cognitive Computing, Educational Data Mining, LINE Official Account, Machine Learning,
Source:	Journal of Theoretical and Applied Information Technology 31^st December 2025 -- Vol. 103. No. 24-- 2025
Full Text

Title:	REAL-TIME ADAPTIVE TRAFFIC MANAGEMENT USING MACHINE LEARNING AND INTERNET OF THINGS (IOT)
Author:	B SARITHA, Dr. KALYANAPU SRINIVAS, PAVAN KUMAR KOLLURU, SIVA KUMAR NALLA, Dr. DIVVELA SRINIVASA RAO, GARIGIPATI RAMA KRISHNA, Dr. SUVETHA POYYAMANI SUNDDARARAJ
Abstract:	This work discusses an adaptive traffic management approach using IoT sensors and ML to handle urban traffic efficiently. To keep traffic from getting jammed, improve driving times and boost safety, the system changes traffic lights every few seconds using data from cameras, LiDAR and radar sensors. A combination of supervised and reinforcement learning is used to forecast traffic and manage traffic signals. Simulations found that travel time is down by 29.9%, traffic flow efficiency goes up by 19.8%, and collisions are reduced by 78.4% with an adaptive signal system. This system is designed to adapt well to changes in traffic and weather, providing noticeable enhancements for traffic handling. As a result of this research, traffic systems in smart cities should become more efficient, expandable and adaptable, directly reducing city congestion and improving how safely people travel. The primary contribution of this research lies in the integration of heterogeneous IoT sensor data with machine learning models to enable real-time, adaptive traffic management. This contribution advances current knowledge by demonstrating a scalable framework that significantly reduces travel time, enhances traffic flow efficiency, and improves road safety in dynamic urban environments.
Keywords:	Adaptive Traffic Management, Internet of Things (IoT), Machine Learning, Traffic Signal Control, Real-Time Optimization, Smart Cities
Source:	Journal of Theoretical and Applied Information Technology 31^st December 2025 -- Vol. 103. No. 24-- 2025
Full Text

Title:	HYBRID VISION TRANSFORMER-CNN FRAMEWORK FOR AUTOMATED RETINAL DISEASE DETECTION FROM FUNDUS IMAGES
Author:	CHANDRA MOULI DARAPANENI, BELLAM SURENDRA BABU, S.V. SATYANARAYANA, L.N.K. SAI MADUPU, V. SUJAY5, NAULEGARI JANARDHAN, G D K KISHORE, KIRAN KUMAR KAVETI, RALLABANDI CH.S.N.P.SAIRAM
Abstract:	Pathologies impacting the retina, such as diabetic retinopathy (DR), glaucoma, and age-related macular degeneration (AMD), contribute significantly to global visual impairment and blindness. The prompt and precise identification of these afflictions through fundus imaging is imperative for facilitating effective interventions and enhancing patient prognoses. This exploration discloses a cutting-edge hybrid deep learning framework that integrates Vision Transformers (ViTs) with Convolutional Neural Networks (CNNs) to enable multi-class classification of retinal disorders retrieved from fundus imaging. The proposed system leverages the localized feature extraction proficiencies of CNNs in conjunction with the global contextual insights provided by transformer-based attention mechanisms. Publicly accessible datasets, namely APTOS, Messidor, and EyePACS, constitute the foundation for both training and evaluation processes. The methodology encompasses sophisticated preprocessing techniques, including contrast enhancement, optic disc segmentation, and extraction of vascular features, aimed at augmenting diagnostic accuracy. This hybrid technique attains a classification accuracy level of 96.3%, in conjunction with a precision statistic of 95.1%, a recall statistic of 96.7%, and an AUC score of 0.982. Statistical validation through ANOVA corroborates the model’s outstanding performance (p < 0.01) when juxtaposed with standalone CNN and ViT architectures. These outcomes underscore the efficacy of intelligent vision-based systems in ophthalmic diagnostics and accentuate their significance within the realms of soft computing, AI-enhanced healthcare, and automated disease screening technologies
Keywords:	Retinal Disease Detection, Fundus Imaging, Deep Learning, Convolutional Neural Networks (CNN), Medical Image Analysis, Diabetic Retinopathy, Age-related Macular Degeneration.
Source:	Journal of Theoretical and Applied Information Technology 31^st December 2025 -- Vol. 103. No. 24-- 2025
Full Text

Title:	AN OPTIMIZED DEEP LEARNING FRAMEWORK FOR ACCURATE AND EFFICIENT FACIAL EMOTION RECOGNITION
Author:	MUSHIKA SHYLAJA, PROF. M SHESHIKALA
Abstract:	This paper presents an end-to-end framework for facial emotion recognition (FER) that is expressly designed for edge deployment where compute, memory, and power are limited. Starting from a MobileNetV2 backbone, we integrate hybrid attention to emphasize semantically meaningful regions (eyes, brows, mouth) while suppressing background noise, yielding features that are both compact and interpretable. To address domain shift across datasets and capture conditions, we adopt a dual-attention domain-adaptation stage that stabilizes representations without expensive target-label supervision. A landmark-guided pruning step then removes redundant filters tied to low-saliency areas, preserving expression-relevant structure while reducing model size. The resulting pipeline balances four goals often treated in isolation accuracy, efficiency, robustness, and accountability. Qualitative visualizations show that attention concentrates on discriminative facial musculature, while stressor analyses (occlusion, illumination, pose/blur) indicate resilient behaviour. The operational profile demonstrates real-time throughput with lower latency and energy per inference, enabling practical deployment on low-power platforms without retraining. Beyond performance, we discuss fairness and privacy considerations and outline auditing hooks that make model decisions easier to inspect. Overall, the framework advances FER from accuracy-only optimization to a deployable, trustworthy, and resource-efficient solution.
Keywords:	Facial Emotion Recognition, Attention Mechanisms, Domain Adaptation, Model Quantization, Edge Deployment
Source:	Journal of Theoretical and Applied Information Technology 31^st December 2025 -- Vol. 103. No. 24-- 2025
Full Text

Title:	MULTIMODAL AI-DRIVEN EARLY DETECTION OF PARKINSON’S DISEASE USING NEURO-MOTOR BIOMARKERS FROM KEYSTROKE DYNAMICS AND HANDWRITING ANALYSIS
Author:	MRS. G.MANI , DR.S.V.G.REDDY
Abstract:	Parkinsons Disease (PD) is an evolving neurodegenerative condition that affects the fine motor control capabilities and thus early diagnosis is important to ensure that there is timely intervention. This research paper introduces a multimodal artificial intelligence (AI) model that combines the keystroke dynamics and the handwriting analysis in detecting early neuro-motor abnormalities linked to PD. The supervised machine learning models of Support Vector Machine (SVM), Random Forest (RF), and Gradient Boosting (GB) were used to analyze the temporal keystroke dataset (Tappy Keystroke Data) with a maximum classification accuracy of 96.12%. At the same time, spatial samples of handwriting (spiral and wave drawings) were handled in the Convolutional Neural Networks (CNNs), and the resultant accuracy was 86.7 and 83.3 accordingly. The association of time and space biomarkers shows that multimodal AI analysis may be more effective than single-modality systems providing a rather inexpensive, non-invasive, and scalable diagnostic tool to support the detection of PD in its early stages. The study forms a base on the development of multimodal detection systems that require the use of other biomarkers including voice and ocular movement to increase predictive accuracy.
Keywords:	Parkinson s Disease, Keystroke Dynamics, Handwriting Analysis, Artificial Intelligence, Neuro-Motor Biomarkers, Machine Learning, Convolutional Neural Networks, Early Detection.
Source:	Journal of Theoretical and Applied Information Technology 31^st December 2025 -- Vol. 103. No. 24-- 2025
Full Text

Title:	ADAPTIVE SLANG AND LANGUAGE MODELING FOR MINING INFORMAL WEB CONTENT
Author:	AMBEDKAR KANAPALA, M SANDEEP, JAKKA TEJA, ASHWINI MANIKONDA, BHASKAR MEKALA,S MAHIPAL
Abstract:	This graph highlights the accuracy comparison of three models—BERT, GPT, and the proposed model—across three datasets: Twitter, Reddit, and Mixed-Language. The proposed model consistently outperforms the baseline models, achieving the highest accuracy across all datasets due to its adaptive slang and language modeling framework. On the Twitter dataset, which contains heavy slang usage, the proposed model achieves an accuracy of 86.1%, significantly higher than BERT (78.4%) and GPT (80.1%). This result demonstrates the model’s superior ability to handle informal and slang-heavy language. On the Reddit dataset, which includes a mix of formal and informal content, the proposed model achieves 84.5% accuracy, showcasing its adaptability to diverse communication styles. Similarly, in the Mixed-Language dataset, the proposed model achieves 81.2% accuracy, highlighting its effectiveness in processing code-switching and multilingual content. Overall, the results indicate that the proposed framework excels in handling informal, noisy, and multilingual data, outperforming traditional models across all tested scenarios
Keywords:	Adaptive Language Modeling, Slang Detection, Informal Web Content, Noisy Data, Continual Learning, Self-Supervised Learning, Sentiment Analysis
Source:	Journal of Theoretical and Applied Information Technology 31^st December 2025 -- Vol. 103. No. 24-- 2025
Full Text

Title:	USING ARTIFICIAL INTELLIGENCE FOR DIGITAL FINANCIAL TRANSACTION MONITORING
Author:	HASSAN ALI AL-ABABNEH, OLEH ONYSHKO2, ANNA PANCHENKO, MARYNA SHASHYNA, OLEKSANDR BARINOV
Abstract:	Although numerous studies have addressed fraud detection in financial systems, there remains a significant gap in the literature regarding validated, scalable AI-based models and hybrid automation schemes for financial transaction monitoring. Existing research lacks comprehensive simulation-based evaluations of transition strategies from human-centric oversight to AI-enhanced processes. This study addresses this gap by identifying optimal AI models and simulating hybrid deployment schemes, thereby generating new knowledge for automated fintech oversight. The following methods were used: focus on the subject of the study; normalization of an unbalanced empirical dataset using Z-score; preparation of a training and validation dataset; comparative analysis of models for detecting anomalies in financial transactions; modelling the risk of committing fraud during the monitoring of financial transactions. It was found that AI models, in particular Random Forest (80% Recall, 99.98% Specificity, 99.95% Accuracy, 0.05% Error, κ = 0.842, Time Cost = 26 s), have significant potential in fintech for detecting fraudulent transactions, and the most effective approach is a hybrid interaction of AI and specialists. The academic novelty of the study is the modelled transition from human monitoring of financial transactions to automation through hybrid schemes, which allows AI to integrate the best practices of specialists. Further research should focus on expanding the range of AI models, using raw datasets without PCA, and improving hybrid interaction schemes in fintech.
Keywords:	Fintech, Fraudulent Transactions, Fraud Detection, Machine Learning, Classification, Empirical Dataset, AI Models.
Source:	Journal of Theoretical and Applied Information Technology 31^st December 2025 -- Vol. 103. No. 24-- 2025
Full Text

Title:	LINKED MOTION ESTIMATION METHOD BASED FEATURE VECTOR WITH PSO INTEGRATED MEAN SHIFT TECHNIQUE FOR VIDEO STEGANOGRAPHY
Author:	SAMEERUNNISA SHAIK , JABEZ J
Abstract:	Video steganography is the practice of concealing a message in a video file in order to transmit it invisibly. For secure online and all communication, the ability to use digital steganography to hide communications is essential. Security applications rely on steganography, which is embedded in video frames and films, to protect sensitive information. In order to reduce data transit and storage requirements, most social media networks use lossy video transcoding. Most video steganography systems cannot be used for secure social media-based communication due to the process of video transcoding. In order to establish dependable secret communication on social media platforms, a robust video steganography model is proposed using the extracted featured from the video to counteract the effects of video transcoding. This research presents a novel method for object-based video steganography, which involves encoding secret data into the motion vectors of objects in motion. Consequently, the objects present in each frame are identified using the mean shift technique. The Motion Vector (MV) of every item is restored to within a quarter of a pixel using a method that estimates motion from B and P frames. Determining the ideal storage-to-video-quality ratio for the chosen motion vectors and associating them with the object requires defining a threshold value. For this reason, we only take into account the motion vectors whose values exceed the threshold. By comparing the magnitude that has been achieved to the threshold value, an action vector is chosen that is larger than the threshold value. For any moving object, the correct vectors are always chosen. Hence, the small motion vectors that didn't come from a geographically significant area are removed. The hidden information is encoded in a quarter of the horizontal and vertical components of the motion vectors that are being tracked. The proposed model makes use of Particle Swarm Optimization (PSO) for updating the feature set for accurate motion detection. In this research a Linked Motion Estimate Method with PSO Integrated Mean Shift Technique using the Weighted Feature Vector (LMEM-PSO-MST-WFV) is proposed for video steganography for secure transmission. The statistics useful for steganography may be affected by changes made to the motion vector during the information embedding process. The proposed model performance in region of motion based feature vector for video steganography is compared with the traditional model and the results represent that the proposed model performance is high.
Keywords:	Video Steganography, Lossy Video Transcoding, Motion Estimate Method, Mean Shift Technique, Feature Vector, Secure Transmission.
Source:	Journal of Theoretical and Applied Information Technology 31^st December 2025 -- Vol. 103. No. 24-- 2025
Full Text

Title:	5G AND BEYOND ADVANCING ULTRA LOW LATENCY COMMUNICATION FOR THE NEXT GENERATION OF AUTONOMOUS SYSTEMS
Author:	DR. L.K. SURESH KUMAR , DR. A. PRASHANTHI , S. SURYANARAYANA , DR. ROSHINI L , JOHN T MESIA DHAS , DR.Y. MADHAVILATHA , P. MUTHUKUMAR
Abstract:	The arrival of 5G has been a radical shift in the telecom industry, offering lightning-fast ultra-low latency capabilities crucial for the up-and-coming waves of autonomous technologies such as self-driving cars, drones, and even other industrial robots. These systems depend primarily on real-time communication and decision-making, enabling autonomous and safe operation. Next Generation Wireless Networks such as 4G and 5G inherit plenty of limitations in addressing the ultra-low latency needs of dynamic autonomous systems, and this paper discusses how edge computing, network slicing, AI-remove optimization, and so on could be an aid to 5G to go ahead in reducing latency and increasing reliability. Specifically, our experiments show that for our hybrid communication framework comprising 5G, edge computing, and dynamic network slicing, we observe a latency of 2.3ms which is notably lower than that of a basic 4G system (50.1ms) and a basic 5G system (20.5ms). The framework also obtained 95.2 Mbps throughput, 99.8% reliability, and shortened time to complete all tasks to 10.2 seconds. The hybrid framework proposed in this article not only outperforms on several vital metrics but also provides a durable approach to satisfy the communication demands of autonomous systems − one that can be further enhanced by optimization of the AI algorithms to determine dynamic resource allocation, as well as the exploitation of 6G to provide reduced latency and intelligent communication capabilities.
Keywords:	5G Technology, Autonomous Systems, Ultra-Low Latency Communication, Edge Computing, Network Slicing
Source:	Journal of Theoretical and Applied Information Technology 31^st December 2025 -- Vol. 103. No. 24-- 2025
Full Text

Title:	DEVELOPMENT OF A RULE-BASED ENERGY MANAGEMENT STRATEGY FOR HYBRID ELECTRIC VEHICLES WITH INTEGRATED HYBRID ENERGY STORAGE SYSTEM
Author:	SUNKA DIVYA , Dr. D. KIRAN KUMAR
Abstract:	Electric vehicles present a compelling alternative to conventional gasoline-powered vehicles, leveraging efficient energy management techniques and showcasing promising prospects for the continued expansion of Hybrid Electric Vehicles (HEVs). The energy management system (EMS) holds a crucial role in HEVs, serving to extend the vehicle's driving range while concurrently lowering costs. This paper introduces an energy management strategy and optimization approach for HEVs equipped with a hybrid energy storage system comprising with Li-ion battery pack and ultracapacitor (UC) pack, implemented through a Fuzzy Logic Controller (FLC). The primary objectives include enhancing battery performance metrics such as state of charge, driving range, and battery lifespan. A parametric comparison is provided with PID controller. Electric vehicle (EV) powered by Batteries offer high energy density, low environmental impact, and durable performance. The integration of both batteries and ultracapacitors is employed to enhance power characteristics. Batteries are used like a primary energy source because these are having high energy density and Ultra capacitors are used for high power density. Simulink serves as a tool for simulating the proposed system's performance, aiding in the determination of the most suitable intelligent system for achieving power efficient operation of the HEV. A comparative analysis of the results from both control techniques is conducted, ultimately recommending the Fuzzy controller for effective energy management in electric vehicles.
Keywords:	Energy storage systems (ESS), Fuzzy logic controller (FLC), Hybrid Electric Vehicle (HEV), Integrated system, Ultracapacitor (UC).
Source:	Journal of Theoretical and Applied Information Technology 31^st December 2025 -- Vol. 103. No. 24-- 2025
Full Text

Title:	ANALYSING HUMOUR AND SENTIMENT IN FAMILY GUY DIALOGUE USING NLP AND MACHINE LEARNING
Author:	PRATHIMA TIRUMALAREDDY , ESWAR REDDY KETHIREDDY , SIRISHA ALAMANDA , LAKSHMI SREENIVASA REDDY DIRISINAPU
Abstract:	Trends and sentiments in the dialogue of Family Guy, a famous animated television show were analyzed. Using the Linear Discriminant Analysis (LDA) machine learning algorithm, prevalent topics and sentiments in the dialogue are identified. The relationship between humour in the dialogue and season-wise ratings is examined to provide insight into what makes Family Guy appealing to its audience. Results suggest that the show's humour, use of pop culture references, and social commentary contribute to its success. Additionally, sentiment analysis reveals that the show's humour is often characterized by sarcasm and irony. Four lexicons are used to identify positive and negative sentiment in the Family Guy dialogue. Three of the lexicons, Affin, Bing, and NRC, are existing, while the fourth lexicon is curated to analyse humour. The sentiment of the dialogue in each dataset is analysed, and results show that the sentiment varies depending on the characters and situations involved. The show often uses humour to address serious social issues. A comprehensive analysis of the dialogue in Family Guy and its implications for media and society is provided.
Keywords:	Sentiment Analysis, Dialogue, Lexicon, Humour, Linear Discriminant Analysis-LDA, Family Guy Dataset
Source:	Journal of Theoretical and Applied Information Technology 31^st December 2025 -- Vol. 103. No. 24-- 2025
Full Text

Title:	ADAPTIVE HYBRID VISION TRANSFORMER FOR PRECISE SEGMENTATION OF MANGO LEAF DISEASES
Author:	NURUL AKMAR AZMAN, FATIMAH KHALID, PUTERI SUHAIZA SULAIMAN,ZAINAL ABDUL KAHAR
Abstract:	Early and accurate detection of leaf diseases is critical for sustaining crop health and preventing yield loss in modern agriculture. However, existing deep learning segmentation models often struggle to balance local feature extraction with global contextual understanding. This study introduces an adaptive hybrid architecture that integrates U-Net’s precise localization with a Vision Transformer (ViT) encoder enhanced by adaptive positional embeddings, which dynamically preserve spatial relationships and adapt to varying leaf morphologies. A tailored preprocessing pipeline combining context-aware brightness normalization and HSV-based background removal further improves symptom visibility under diverse lighting conditions. Using the MangoLeafBD dataset (2,500 annotated images across five classes) to validate the framework, the proposed model achieves 99.71% accuracy and a mean IoU of 0.973, surpassing existing methods by 0.78–8.13% while maintaining efficiency on mid-range GPUs. This study establishes a scalable benchmark for agricultural image segmentation, advancing the architecture of hybrid transformer with CNN integration for field-level disease detection.
Keywords:	Vision Transformers, U-Net, Semantic Segmentation, Adaptive Positional Embeddings
Source:	Journal of Theoretical and Applied Information Technology 31^st December 2025 -- Vol. 103. No. 24-- 2025
Full Text

Title:	MULTI-HOP COMMUNICATION IN WIRELESS SENSOR NETWORK
Author:	RAKESH RANJAN, VAISHALI SINGH, HITENDRA SINGH
Abstract:	In this research, the author proposes a novel multi-hop communication framework for Wireless Sensor Networks (WSNs) aimed at enhancing clustering efficiency and securing data transmission against intruder-based attacks. The core contribution lies in an optimized Cluster Head (CH) selection algorithm combined with a lightweight Intrusion Detection Mechanism (IDM), ensuring both energy efficiency and secure communication. Simulations were conducted using NS-3 (Network Simulator-3), where 150 sensor nodes were deployed over a 100×100 m² area. Performance was evaluated over 500 rounds using multiple performance metrics. The proposed Secure Energy-aware Multi-hop Adaptive Clustering (SEMAC) model demonstrated a 28.6% improvement in clustering efficiency compared to the traditional LEACH protocol, along with a 17.4% reduction in energy consumption per node per round. In terms of security, SEMAC achieved a 96.2% intrusion detection rate, a false positive rate of only 3.5%, and an attack detection accuracy of 95.8%, highlighting its robustness against common WSN attacks. The model converged efficiently within 80 training epochs, reducing the overhead of repeated clustering. The prediction module for node behaviour and trust scores exhibited low error margins, with a Mean Absolute Error (MAE) of 0.046 and a Mean Squared Error (MSE) of 0.0089. Additionally, the system maintained a high Packet Delivery Ratio (PDR) of 93.7%, ensuring reliable data transmission even under security threats and energy constraints. These results validate the effectiveness of the SEMAC model in enhancing the operational lifespan, reliability, and resilience of WSNs, making it suitable for deployment in mission-critical IoT and environmental monitoring applications.
Keywords:	Wireless Sensor Networks, Clustering Efficiency, Intrusion Detection Mechanism, SEMAC, MAE, MSE, NS-3.
Source:	Journal of Theoretical and Applied Information Technology 31^st December 2025 -- Vol. 103. No. 24-- 2025
Full Text

Title:	APPLICATION OF THE DEMPSTER SHAFER METHOD IN AN EXPERT SYSTEM TO DETECT ONLINE GAME ADDICTION
Author:	MARYANA, NURDIN, TAUFIQ, NURUL WILDA, MUHAMMAD FIKRY, ARNAWAN HASIBUAN
Abstract:	The increasing popularity of online games among college students has given rise to various problems, one of which is online game addiction. The impacts of online game addiction include academic, social, and psychological aspects in students. The purpose of this study is to build an expert system to diagnose online game addiction using Dempster Shafer, using questionnaire data distributed to 264 students as a dataset and 211 students as training data. The Dempster Shafer method uses expert confidence values to determine how much a symptom can influence the likelihood of a case occurring. Research on this expert system used 16 symptoms and 3 levels of addiction: mild addiction, moderate addiction, and severe addiction. The results of the study using 53 test data showed the Dempster Shafer method has an accuracy rate of 73%, resulting in 39 moderate addictions. With this system model, it is hoped that it can provide information and solutions for people who have experienced addiction so that prevention and recovery can be carried out to avoid fatal consequences.
Keywords:	Online Games, Expert System, Dempster Shafer Method, Game Addiction Symptoms.
Source:	Journal of Theoretical and Applied Information Technology 31^st December 2025 -- Vol. 103. No. 24-- 2025
Full Text

Title:	AUDIO-VISUAL DEEPFAKE DETECTION WITH CROSS MANIPULATION EVALUATION ON FAKEAVCELEB
Author:	REHAM MOHAMED ABDULHAMIED, MONA M. NASR, FARID ALI MOUSA, SARAH NAIEM
Abstract:	Deepfake technologies have advanced rapidly in recent years, enabling highly realistic manipulations of both audio and video. While such technologies offer creative potential, they also present major risks in misinformation, fraud, and privacy violations. This paper explores the problem of detecting deepfake content using the FakeAVCeleb dataset, which provides both authentic and manipulated audio-video samples. We present experiments using state-of-the-art audio models (AASIST, RawNet2, ECAPA-TDNN), video models (Vision Transformers and SyncNet/Wav2Lip for lip-sync consistency), and multimodal fusion approaches. In particular, we evaluate robust detection under cross-manipulation scenarios, where models are tested on manipulation types unseen during training. Our results highlight the performance drop in cross-manipulation settings, emphasizing the importance of robust multimodal fusion. Fusion methods achieved improved generalization, indicating that combining complementary cues across modalities is key to resilient deepfake detection.
Keywords:	Deepfake Detection, FakeAVCeleb Dataset, Audio-Visual Forensics, Cross-Manipulation Evaluation, Video Manipulation, Audio Deepfake. Visual Deepfake Introduction.
Source:	Journal of Theoretical and Applied Information Technology 31^st December 2025 -- Vol. 103. No. 24-- 2025
Full Text

Title:	MRI BASED ALZHEIMER’S DISEASE DETECTION AND STAGING USING ATTENTION-GUIDED HYBRID ENSEMBLE LEARNING
Author:	VINUKONDA EMANUEL RAJU , B.N. JAGADESH
Abstract:	The prompt and accurate identification of Alzheimer s Disease (AD) is essential for guiding effective treatment strategies. However, this task is challenging due to the similar structural brain features present across different stages of the disease and the uneven distribution of classes in medical imaging datasets. This study introduces a hybrid deep learning framework aimed at improving the multi-class classification of Alzheimer's disease based on MRI scans. The technique begins with image preparation utilizing CLAHE and elastic deformation to enhance contrast and variability. Brain structures are then delineated using an Attention U-Net, enabling the model to emphasize anatomically significant regions. Feature collection is obtained from the EfficientNetV2B0 architecture and subsequently subjected to dimensionality reduction through CatBoost based feature selection, which retains the most significant properties. A Deep Neural Network (DNN) utilizing restructured attention is trained on a balanced dataset, augmented through SMOTE to rectify class imbalance. The final prediction is generated by integrating outputs from CatBoost, XGBoost, and the DNN through an ensemble meta-learner utilizing attention-based integration. Experimental validation on the OASIS dataset achieved an accuracy of 95.47%, exceeding current benchmark models. The suggested method exhibits significant generalization and sensitivity to minority classes, underscoring its potential applicability in clinical diagnostic procedures.
Keywords:	Alzheimers Disease Classification, Attention U-Net, EfficientNetV2B0, Ensemble Learning, MRI Brain Imaging.
Source:	Journal of Theoretical and Applied Information Technology 31^st December 2025 -- Vol. 103. No. 24-- 2025
Full Text

Title:	A DEEP TREE LIQUID STATE AND GREEDY STACKED AUTOENCODER FRAMEWORK FOR THYROID DISEASE PREDICTION
Author:	Mrs. R. VANITHA , Dr. K. PERUMAL
Abstract:	Thyroid disease is a medical state of affairs in which the thyroid gland potentiality to be in action is affected. The thyroid gland being an endocrine organ positioned in front of neck that induces thyroid hormones and advances via the bloodstream to aid control a variety of other organs. On one hand when body induces too much thyroid hormone, hyperthyroidism is said to be developed whereas on the other hand insufficient hormone production results in hypothyroidism. Deep Learning (DL) algorithms, though analyze data in an automatic manner and constructing analytical models according to DL can also be utilized for thyroid disease diagnosis. Feature extraction is essential in DL to improve efficiency where features are extraction based on their significance. It means that the most important elements of the DL model have been extracted for prediction by discarding irrelevant features that have nothing to do with the disease prediction. In this work a method called Deep Tree Liquid State and Greedy Stacked Autoencoders (DTLS-GSA) based feature extraction for thyroid disease prediction. The DTLS-GSA based feature extraction for thyroid disease prediction is split into two sections, namely, efficient data handling and feature extraction. First, data handling employs thyroid raw datasets are performed using Deep Tree Liquid State Machine based architecture. The objective of employing the Deep Tree Liquid State Machine is the potential in handling data that are highly skewed without the loss of original information. Second, with efficient handling of data, Greedy Layerwise-Stacked Denoising Sparse Autoencoders is applied to extract the features required for thyroid disease detection. Also hereby employing both forward propagation (i.e., inputting sample instances) and backward propagation (i.e., fine tuning weights) therefore minimizes loss function convergence at an early stage, therefore improving overall precision and recall rate of extracted features. Experimental evaluation will be carried out using Thyroid Disease dataset with different factors such as, precision, recall, accuracy and training time.
Keywords:	Thyroid Disease Prediction, Deep Learning, Deep Tree Liquid State Machine, Greedy Layerwise, Stacked Denoising, Sparse Autoencoders
Source:	Journal of Theoretical and Applied Information Technology 31^st December 2025 -- Vol. 103. No. 24-- 2025
Full Text

Title:	AN HHO OPTIMIZED CNN VIT FRAMEWORK FOR CERVICAL CELL CLASSIFICATION
Author:	REMYA R , KUMUDHA RAIMOND
Abstract:	Accurate identification and classification of cervical cells is crucial for the early identification and treatment of cervical cancer. In image classification, the ability to capture both local texture and global contextual features is crucial for improving model performance. Convolutional Neural Networks (CNN) are effective in extracting local features using hierarchical filters whereas Vision Transformers (ViT) are good at capturing long range dependencies using self-attention mechanisms. This study presents a hybrid method that combines the features extracted by ResNet50 and ViT architectures to integrate their complementary strengths. The process incorporates Harris Hawks Optimization (HHO) to extract discriminative features and reduce their dimensionality. Finally, an SVM was used to classify the optimized features. The method was evaluated on two benchmarked datasets, SIPaKMeD and Mendeley, achieving impressive accuracies on both datasets. The results demonstrate that the hybrid method significantly enhances cervical cell classification and can support more reliable clinical decisions in cervical cancer detection.
Keywords:	Cervical cancer, ResNet 50, ViT, Harris Hawks Optimization (HHO), SVM
Source:	Journal of Theoretical and Applied Information Technology 31^st December 2025 -- Vol. 103. No. 24-- 2025
Full Text

Title:	A BIOMETRIC AND SECURITY REINFORCED TECHNIQUE WITH HYBRID DEEPFAKE DETECTION
Author:	VENKATESWARLU SUNKARI, A. SRINAGESH
Abstract:	Biometric authentication systems, such as facial recognition and voice identification, are increasingly used for secure access across devices and services. However, the rapid advancement of deepfake technology—powered by generative models like GANs and autoencoders—poses a significant threat by enabling realistic synthetic media that can spoof biometric systems. These manipulated inputs can bypass traditional security mechanisms, undermining the reliability of identity verification. This paper proposes a Hybrid DeepFake Detection Framework to reinforce biometric security by integrating spatial, temporal, frequency-based, and physiological signal analysis. The system leverages Convolutional Neural Networks (CNNs) for image-based forgery detection, Long Short-Term Memory (LSTM) networks for temporal consistency, and Fourier Transform techniques for frequency-domain anomaly detection. Additionally, it incorporates micro-expression and remote photoplethysmography (rPPG) analysis for detecting liveness. Experiments conducted on benchmark datasets—FaceForensics++, Celeb-DF, and DFDC—demonstrate that the hybrid model achieves superior performance compared to baseline models, with an accuracy exceeding 92% and real-time inference capabilities. The system also exhibits strong generalization to unseen deepfake methods and resilience against adversarial perturbations. Compared to existing single-model approaches, our hybrid system significantly improves detection accuracy, robustness, and deployability, providing a reliable solution to counter evolving deepfake threats in biometric authentication systems.
Keywords:	Biometric Authentication, Deepfake Detection, Hybrid Framework, Convolutional Neural Networks , FaceForensics++
Source:	Journal of Theoretical and Applied Information Technology 31^st December 2025 -- Vol. 103. No. 24-- 2025
Full Text

Title:	DP-MCR: A BIO-INSPIRED MULTI-CONSTRAINT ROUTING PROTOCOL FOR DELAY-RESILIENT AND SECURE COMMUNICATION IN AGRICULTURAL IOT NETWORKS
Author:	V. S. JAGADEESWARAN , DR. K. SANTHI
Abstract:	The Internet of Agricultural Things (IoAT) has become central to precision farming, yet routing in these networks remains constrained by limited energy, unstable connectivity, and dynamic traffic conditions. To address these challenges, this research proposes Danaus Plexippus-Based Multi-Constraint Routing (DP-MCR), a biologically inspired framework that adapts routing behaviour through natural migration strategies of monarch butterflies. DP-MCR integrates swarm-based clustering to balance communication loads, gradient-driven constraint weighting to refine routing decisions, and nectar-inspired path optimization to select stable, energy-rich relays. A gliding-based multi-hop mechanism minimizes retransmissions, while migration-guided load redistribution and self-healing strategies maintain resilience in the face of node failures and congestion. Real-time, quality-of-service-aware scheduling further ensures the timely delivery of critical agricultural data. Simulation analysis demonstrates that DP-MCR reduces energy consumption by over 30%, lowers the average delay by nearly 20%, maintains packet delivery ratios above 78%, and improves throughput beyond 290 kbps, while achieving the lowest packet loss across varying densities. By embedding adaptive swarm intelligence into routing decisions, DP-MCR delivers an energy-aware, delay-resilient, and fault-tolerant solution tailored to the operational requirements of agricultural ecosystems in 2025.
Keywords:	IoT, IoAT, Agriculture, WSN, Routing, Optimization, Swarm Intelligence.
Source:	Journal of Theoretical and Applied Information Technology 31^st December 2025 -- Vol. 103. No. 24-- 2025
Full Text

Title:	BLOOD-BASED BIOMARKERS FOR EARLY PREDICTION OF KNEE BONE CANCER (EPKC): A PREDICTIVE MODELING TECHNIQUE FOR OSTEOSARCOMA AND CHONDROSARCOMA DIAGNOSIS
Author:	G.JANAKI, D.UMANANDHINI
Abstract:	Knee bone cancers such as Osteosarcoma and Chondrosarcoma are challenging to diagnose in the early stages because their symptoms often resemble those of non-cancerous conditions like Osteoarthritis. In this study, nine blood based biomarkers – CRP, IL-6, TNF-α, COMP, MMP-3, VEGF, ESR, LDH, and ALP – were used in a machine learning based diagnostic model to classify patients into four categories: Chondrosarcoma, Osteosarcoma, Normal, and Misclassified Cancer. The model was trained on a dataset of 5119 patient records using a Random Forest Classifier optimised through hyperparameter tuning. It is being hypothesized that a biomarker-driven, hyperparameter-optimized Random Forest would distinguish between Osteosarcoma, Chondrosarcoma, and normal cases at a very high accuracy. It achieved a classification accuracy of 0.95. Feature important analysis identified LDH, ALP, and CRP as the most predictive biomarkers. Receiver operating characteristic (ROC) analysis further confirmed strong class separability, with an area under the curve (AUC) of 0.95. The dataset used in this study was delivered from clinical data between collected between 2022 and 2025, containing blood biomarker profiles with confirmed diagnostic labels. These findings demonstrate the potential of integrating multi-biomarker data with artificial intelligence to support early and non-invasive detection of knee bone cancers, which could aid clinicians in timely diagnosis and treatment planning.
Keywords:	Osteosarcoma, Chondrosarcoma, Random Forest Classifier, Hyperparameter tuning, Non-invasive Diagnosis.
Source:	Journal of Theoretical and Applied Information Technology 31^st December 2025 -- Vol. 103. No. 24-- 2025
Full Text

Title:	PERFORMANCE ASSESSMENT OF NHPP SOFTWARE RELIABILITY MODELS APPLYING SHAPE PARAMETER OF WEIBULL-TYPE LIFE DISTRIBUTIONS
Author:	HYO JEONG BAE
Abstract:	This study was conducted with the aim of evaluating the performance properties issue of NHPP-based software reliability models parameterized by the shape parameter of the Weibull-type life distributions. For this purpose, failure time data collected from actual software fault occurrences were utilized, and the model parameters were estimated using the MLE approach. The reliability performance of the proposed models was then assessed through several analytical approaches, including model efficiency evaluation using MSE and R², the degree of accuracy in prediction against observed values using m(t), instantaneous failure rate analysis by λ(t), and reliability prediction based on future mission times. The results revealed that the Inverse-Exponential model demonstrated superior efficiency compared with the other models. Accordingly, this study provides a novel analysis of performance issues in reliability models governed by the shape parameter of the Weibull distribution, an area insufficiently addressed in prior studies, and offers basic reliability properties that can be helpful to developers in the preparation stage.
Keywords:	Exponential, Inverse-Exponential, NHPP, Rayleigh, Property Assessment.
Source:	Journal of Theoretical and Applied Information Technology 31^st December 2025 -- Vol. 103. No. 24-- 2025
Full Text

Title:	HYBRID DEEP LEARNING APPLICATION FOR SOLAR IRRADIANCE FORECASTING MODEL
Author:	SAUMYA MISHRA , DEEPENDRA PANDEY, SAURABH BHARDWAJ
Abstract:	Secure scheduling, cost-effective power system operations, and the reduction of technical and financial risks for the electricity market all depend on accurate and trustworthy solar irradiance forecasting (SIF). This paper introduces a new hybrid, data-driven deep learning scheme called SIF. Various geographical and meteorological datasets are preprocessed, and then redundant variables are removed using feature selection (FS) and Grey wolf optimization (GWO-VMD)-enhanced variational mode decomposition. The refined data used for predicting the subsequence of solar irradiance by the BiLSTM-Attention network. Lastly, kernel density estimation for Gaussian kernel (KDE-Gaussian) is used to define prediction intervals for different confidence levels. The comparison shows that the FS-GWO-VMD-BiLSTM-Attention scheme performs better, reducing MAE/MAPE/MSE by 93%, 86%, and 98%, respectively to BPNN model. With narrower range, higher coverage, and deeper information of interval, the KDE-Gaussian approach produces outstanding coverage–width criterion (CWC) results that precisely reflect fluctuations in SI.
Keywords:	Solar Irradiance Forecasting, Bidirectional Long Short-Term Memory, Data-Driven Modeling, Interval Forecasting, Attention Mechanism
Source:	Journal of Theoretical and Applied Information Technology 31^st December 2025 -- Vol. 103. No. 24-- 2025
Full Text

Title:	APPLICATION OF CLOUD COMPUTING TECHNOLOGIES IN DIGITAL MARKETING MANAGEMENT OF COMPANIES
Author:	NOOR AHMAD ALKHUDIERAT, HASSAN ALI AL-ABABNEH , HAMZA MOHAMMED MASHAQBEH , FAYEZ M. AL-KHAWALDEH , JAMEEL AHMAD KHADER , JAFAR ABABNEH
Abstract:	The active spread of cloud computing technologies is radically changing the organization of digital marketing in companies. With large volumes of data, multichannel communication and the need for prompt personalization, traditional infrastructures are no longer effective. A scientific and practical problem arises: how to objectively assess the impact of cloud solutions on digital marketing management. The purpose of this study is to develop and test a methodology for assessing this impact and, based on it, identify the main effects of cloud technologies. The research methodology is based on the construction of a system of digital marketing management performance indicators (data processing speed, infrastructure costs, conversion rate, return on marketing investments) and their comparison before and after the implementation of cloud solutions. Index analysis and normalization of indicators for data comparability were used as a tool; the sample was made up of ten international companies that implemented cloud marketing platforms in 2021-2024. Data was collected on the basis of public corporate reporting and expert interviews; calculations were performed using descriptive statistics and benchmarking methods. The obtained results show that cloud computing increases flexibility and speed of decision-making, reduces costs and improves key indicators of marketing activities, which confirms the feasibility of their use. The developed methodology can be used by managers and researchers for systematic evaluation of the effectiveness of cloud technologies in digital marketing.
Keywords:	Cloud Computing, Digital Marketing, Marketing Process Management, IT System
Source:	Journal of Theoretical and Applied Information Technology 31^st December 2025 -- Vol. 103. No. 24-- 2025
Full Text

Title:	SECURING HEALTHCARE RECORDS: A BLOCKCHAIN-BASED APPROACH FOR DATA INTEGRITY AND ACCESS CONTROL
Author:	IKRAM BEN ABDEL OUAHAB , LOTFI ELAACHAK , YASSER ALLUHAIDAN , AND MOHAMMED BOUHORMA
Abstract:	The management of medical records presents significant challenges in the healthcare sector, with traditional centralized systems plagued by vulnerabilities such as cyberattacks, unauthorized access, and data breaches. These issues compromise the confidentiality, integrity, and accessibility of sensitive patient information, further exacerbated by the increasing complexity and volume of healthcare data. This paper proposes a blockchain-based framework leveraging Ethereum’s decentralized architecture and the InterPlanetary File System (IPFS) for distributed storage to address these challenges. Smart contracts, implemented in Solidity, automate critical processes such as patient registration, access control, and record updates, ensuring traceability and eliminating intermediaries. Sensitive medical data is encrypted before storage on IPFS, with only cryptographic hashes stored on-chain to reduce storage overhead while maintaining data integrity. Decentralized authentication mechanisms, such as Ethereum wallets and cryptographic keys, enhance security and usability. This approach empowers patients with full control over their data and real-time access management. The system’s scalability, transparency, and patient-centric design provide a robust alternative to traditional models, paving the way for advancements in healthcare data management and beyond.
Keywords:	Blockchain Technology, Medical Record Management, Data Security, Smart Contracts, InterPlanetary File System (IPFS)
Source:	Journal of Theoretical and Applied Information Technology 31^st December 2025 -- Vol. 103. No. 24-- 2025
Full Text

Title:	CONSENSUS LEARNING WITH PATHOPHYSIOLOGY-INFORMED SPECTRAL FEATURES FOR EEG-BASED STROKE AND TBI DIAGNOSIS
Author:	GAO JIALU , NORWATI MUSTAPHA, NORIDAYU BINTI MANSHOR , RAZALI BIN YAAKOB
Abstract:	Traumatic brain injury (TBI) and stroke are distinct yet critical neurological conditions requiring timely and accurate diagnosis. While EEG-based binary classification models often exceed 90% accuracy, clinically relevant three-class classification (TBI vs. stroke vs. normal) remains a significant challenge, with current state-of-the-art methods achieving only 69% accuracy and 0.84 ROC-AUC. This study introduces a novel consensus learning framework that incorporates pathophysiology-informed frequency band decomposition to improve classification performance. Three specialized spectral configurations are employed: a TBI-optimized delta band (0.5–4 Hz), a stroke-sensitive alpha band (8–13 Hz), and a cross-pathology band covering conventional clinical frequencies. Identical TMN-CNN architectures are trained on each spectral view, and an adaptive consensus mechanism integrates predictions via class-specific weighting and confidence-based voting. Evaluated on 9,276 EEG segments from the Temple University Hospital EEG Corpus, the proposed method achieves 76.57% balanced accuracy and 0.911 ROC-AUC, outperforming benchmarks by +0.071 ROC-AUC. Class-specific analysis confirms balanced performance across categories, with a +3.22% gain over the best individual model. These results demonstrate the effectiveness of pathophysiology-guided, multi-view consensus learning for robust EEG-based neurological classification.
Keywords:	EEG Classification, Traumatic Brain Injury, Stroke Detection, Consensus Learning, Multi-Modal Signal Integration)
Source:	Journal of Theoretical and Applied Information Technology 31^st December 2025 -- Vol. 103. No. 24-- 2025
Full Text

Title:	ADAPTIVE DEEP LEARNING FRAMEWORK FOR ROBUS QR CODE WATERMARKING WITH NOISE-AWARE PRESERVATION
Author:	SHIKHA YADAV, SHILPA SHARMA, DR. RAHAMA SALMAN, RUBEENA M. AKHTAR, SAYYADA SARA BANU
Abstract:	Digital watermarking has emerged as an essential technology for copyright protection and content authentication in the era of exponential digital content proliferation. Quick Response (QR) codes, with their inherent error correction capabilities and high information capacity, represent promising watermark carriers that enable machine-readable authentication. However, existing watermarking methods face fundamental challenges in simultaneously achieving robust watermark survival against diverse attacks while maintaining imperceptible embedding quality. Traditional transform-domain techniques suffer from limited adaptability and manual parameter tuning, while recent deep learning approaches lack specialized mechanisms for preserving watermark integrity during noise removal and exhibit insufficient robustness under sophisticated attacks. To address these limitations, this paper proposes a novel deep learning framework for QR code watermarking with three key innovations: (1) an adaptive strength encoder that automatically modulates embedding intensity based on local image characteristics to optimize the imperceptibility-robustness trade-off, (2) a dual-detection denoising module featuring independent watermark and noise detectors that selectively preserve embedded information while suppressing various noise types, and (3) a multi-scale quality-aware decoder with parallel feature extraction paths and confidence estimation for robust QR reconstruction under severe distortions. Extensive experiments on challenging datasets demonstrate superior performance with QR extraction accuracy exceeding 96%, watermark PSNR above 37 dB, and denoising quality surpassing 31 dB, substantially outperforming state-of-the-art approaches.
Keywords:	Digital Watermarking, QR Codes, Deep Learning, Noise-Aware Denoising, Adaptive Embedding
Source:	Journal of Theoretical and Applied Information Technology 31^st December 2025 -- Vol. 103. No. 24-- 2025
Full Text

Title:	MULTI-SCALE FEATURE FUSION IN 3D CNNS FOR AORTIC ANEURYSM DETECTION IN ANGIOGRAPHY SCANS
Author:	Dr.B.KALPANA, Dr.N.SUBHASH CHANDRA, Dr.K.SELVAM, DR.T.VENGATESH
Abstract:	Abdominal Aortic Aneurysm (AAA) is a life-threatening condition characterized by the pathological dilation of the aorta. Early detection through Computed Tomography Angiography (CTA) scans is crucial for preventing rupture, but manual screening is time-consuming and subject to inter-observer variability. While 3D Convolutional Neural Networks (CNNs) are a natural fit for volumetric medical image analysis, they often struggle with the significant scale variation of AAAs, which can range from subtle, localized dilations to extensive, tortuous structures spanning large volumes. This paper introduces MS-FFNet3D, a novel deep learning architecture designed to overcome this challenge through an advanced multi-scale feature fusion strategy. MS-FFNet3D is built upon a 3D encoder-decoder backbone integrated with a Multi-Scale Fusion (MSF) module. This module employs a combination of atrous spatial pyramid pooling (ASPP) and dense connections to extract and aggregate features at multiple receptive fields, enabling the network to capture both fine-grained contextual details for precise boundary delineation and global contextual information for overall aneurysm presence. Evaluated on a curated dataset of 358 CTA scans, our model achieved a Dice similarity coefficient of 91.4% for aneurysm segmentation and an AUC of 0.984 for detection, significantly outperforming standard 3D U-Net (Dice: 87.2%, AUC: 0.942) and V-Net (Dice: 88.1%, AUC: 0.951) baselines. The proposed method demonstrates high sensitivity in detecting small, nascent aneurysms, offering a powerful tool for automated, large-scale screening and precise measurement in clinical settings.
Keywords:	Abdominal Aortic Aneurysm, Computed Tomography Angiography, 3D Convolutional Neural Networks, Multi-Scale Feature Fusion, Medical Image Segmentation, Computer-Aided Diagnosis.
Source:	Journal of Theoretical and Applied Information Technology 31^st December 2025 -- Vol. 103. No. 24-- 2025
Full Text

Title:	FBA SHIELD: ENHANCING PRIVACY PRESERVING VULNERABILITY MANAGEMENT FOR SECURITY BY DESIGN IN MEDICAL IOT ECOSYSTEMS
Author:	J. S. REXON, P. MANIKANDAN, C. SWEDHEETHA
Abstract:	The rapid development of Internet of Things (IoT) in the medical industry has greatly enhanced patient monitoring and diagnosis, although it has also introduced a new and unexplored scale of cybersecurity risks. The traditional methods of vulnerability management in medical IoT ecosystems do not necessarily work well in offering a high level of protection and stringent privacy protection as they are founded on centralized designs that expose sensitive medical data to leakage, manipulation, and attacks. To address these weaknesses, this paper proposes an integrated architecture, FBA SHIELD (Federated Blockchain-Assisted SHIELD), to provide privacy-sensitive vulnerability management based on the security-by-design principles. The suggested model is a federated learning model which is applied to do decentralized anomaly detection, and which is enhanced with blockchain-based encryption and consensus in order to provide a secure model update and prevent data-poisoning. It has been used to implement the framework in Python using the TensorFlow and blockchain libraries and tested on the IoT Medical Devices Cybersecurity Dataset in Kaggle, in order to be reviewed comprehensively against the baseline models. Empirical findings indicate that FBA SHIELD is 97.8% accurate, 96% precise, 95% recall, 96% F1-score, and an AUC of 98.5%, and 4.4 points higher than high-end approaches with a lower computational cost. The blockchain layer also showed a throughput (TPS) of 245 transactions per second and an average latency of 1.8 seconds, which guaranteed efficiency and reliability in the decentralized healthcare settings. In this study FBA SHIELD, which combines federated learning with blockchain to further secure and privacy-sensitive vulnerability management of medical IoT networks. It also brings fresh knowledge on decentralized trust verification, collaborative model training, and optimized efficiency in anomaly detection, which can be used to scale up to privacy-conscious healthcare cybersecurity. The changing attack vectors are also resilient to the system because it has a feedback loop to the model, which is updated in real-time, thereby giving the system adaptive scaling. FBA SHIELD may be used in remote healthcare, smart medical devices, and distributed hospital networks, and the security of sensitive patient information is extremely significant in these scenarios. Its good performance is justified by its potential as an innovative solution, and FBA SHIELD is an effective and deployable system of medical cybersecurity IoT of the new generation.
Keywords:	Blockchain Security, FBA SHIELD, Federated Learning, Medical IoT Cybersecurity, Privacy-Preserving Vulnerability Management
Source:	Journal of Theoretical and Applied Information Technology 31^st December 2025 -- Vol. 103. No. 24-- 2025
Full Text

Title:	MITIGATING ADVERSARIAL STYLOMETRY USING LEF-HT AND C2S-CGReLUNN FOR MULTI-AUTHOR WRITING STYLE ANALYSIS
Author:	RIYA SANJESH, Dr. PAMELA VINITHA ERIC
Abstract:	The analysis of multi-author writing style is necessary to identify the individual author from collaborative text. Most of the prevailing works in research literature to overcome the adversarial stylometry, have paid less or no attention to punctuation usage and geometric shapes of the handwritten text. Therefore, in this work, Cutmix-Shake-Shake Convolutional Gaussian-error Rectifier Linear Unit Neural Network (C2S-CGReLUNN) classifier-based multi-author identification is implemented. Initially, the handwritten images are collected and pre-processed. Later, the edges are detected using the Lanczos Kernel-based Canny Edge Detector (LK-CED). Next, the geometric patterns are analyzed using Logarithmic Exponential Fit Hough Transform (LEF-HT). Further, the stroke movement is detected using the Reflect Padding-based Fast Fourier Transform (RP2FT). Next, the stylometric features are extracted. Meanwhile, the text is recognized, and from this, the punctuation-based features are extracted, followed by the calculation of punctuation density. Also, the vocabulary richness is identified using the Entropy-based Spatial Indexing Bidirectional Encoder Representation from Transformers (ESI-BERT) technique. Finally, the extracted features, punctuation density of the author, and vocabulary richness identified output are given to the C2S-CGReLUNN for author classification. Thus, the multi-author writing style is analyzed effectively with a prediction accuracy of 99.07%.
Keywords:	Multi-Author Writing Style Analysis, Stylometric Features, Deep Learning, Natural Language Processing (NLP), Geometric Character Analysis, Stroke Movement Identification, and Edge Detection, Text Recognition, Convolutional Nueral Network
Source:	Journal of Theoretical and Applied Information Technology 31^st December 2025 -- Vol. 103. No. 24-- 2025
Full Text

Title:	AI-ENHANCED GREEN COMPUTING FRAMEWORK FOR REDUCING DATA CENTER ENERGY FOOTPRINT
Author:	R. SREE CHAITRA, M. HEMANTH, K. SRIKANTH, RAVI KUMAR TATA1, YELISELA RAJESH
Abstract:	The rapid escalation of digital media and the continuous pursuit of high-speed computation have highlighted critical shortcomings in conventional image processing paradigms, particularly concerning scalability, power efficiency, and environmental responsibility. Traditional approaches rely heavily on resource-demanding operations, which contribute significantly to energy usage and carbon emissions, thereby conflicting with the global agenda of sustainable digital ecosystems. Recent research has begun to emphasize sustainable computing directions in visual data analysis, targeting both algorithmic efficiency and hardware optimization to minimize energy overheads. Yet, insights from eco-conscious design in image processing remain fragmented and underexplored. This work addresses the pressing requirement for environmentally aligned image analysis by reviewing existing practices, examining their limitations, and identifying the sustainability trade-offs inherent in current solutions. Furthermore, we propose an optimized framework for feature extraction and interpretation that balances computational capability with reduced energy expenditure. Experimental assessments against conventional baselines confirm the framework’s ability to deliver reliable processing outcomes while achieving notable reductions in energy demand. In addition, this work contributes a unified, data-driven framework integrating key green-computing techniques—virtualization, cooling optimization, hardware–software efficiency improvements, and renewable-energy utilization—to systematically reduce energy consumption in data centers. Unlike studies that investigate these techniques independently, this study presents a combined optimization approach supported by real energy-consumption modeling. A hybrid machine-learning model (SVM+RF+XGBoost) is also developed to accurately predict data-center energy usage, enabling proactive energy-management strategies. Experimental results demonstrate improved prediction accuracy and a measurable reduction in overall energy demand, providing new insights into sustainable and environmentally conscious data-center operations.
Keywords:	Sustainable Image Processing, Energy-Conscious Algorithms, Green Computation, Low-Power Visual Analytics, Resource-Aware Systems
Source:	Journal of Theoretical and Applied Information Technology 31^st December 2025 -- Vol. 103. No. 24-- 2025
Full Text

Title:	DEEP LEARNING POWERED HYBRID FRAMEWORK FOR MULTICLASS HISTOPATHOLOGY BASED CANCER DIAGNOSIS
Author:	PRIYANKA KHABIYA, PROF(DR.)FIROJ PARWEJ
Abstract:	This work introduces an explainable deep learning approach for classifying lung and colon cancer using histopathological images. Recent improvements in deep learning have significantly enhanced medical image analysis; however, most existing models still suffer from poor interpretability, limited transparency, and weak generalization across tissue types. To address these gaps and enhance clinical trust, this study proposes a transparent and reliable deep learning framework for cancer diagnosis. The framework influences a custom Convolutional Neural Network architecture, enhanced by the Swish activation function to enhance merging and representation learning. This architectural selection enables better feature extraction and convergence compared to traditional activation functions, resulting in an improved overview across histopathological patterns. To improve transparency and model responsibility, explainability techniques, such as Gradient-weighted Class Activation Mapping (Grad-CAM) and Integrated Gradients (IG), are employed, providing visual insights into the regions that contribute to model predictions. Evaluated on a standard histopathology dataset, the proposed model achieves high classification performance with 98.69% accuracy, 98.71% precision, 98.69% recall, and 98.70% F1-Score. The visual attribution maps confirm that the network attends to the diagnostically major tissue regions, reinforcing its reliability in medical image analysis. The significant impact of this study is that it demonstrates the potential of combining an optimized CNN architecture with explainable AI methods to improve diagnostic accuracy and interpretability concurrently. The findings propose that combining advanced CNN design with an explainability technique can bring both accuracy and interpretability in digital pathology applications.
Keywords:	Deep Learning, Histopathology, Lung Cancer, Colon Cancer, SegNet, EfficientNet, VGG16, Explainable AI, Grad-CAM, Integrated Gradients
Source:	Journal of Theoretical and Applied Information Technology 31^st December 2025 -- Vol. 103. No. 24-- 2025
Full Text

Title:	KNOWSENTMIX: LEVERAGING CONTEXTUAL EMBEDDINGS AND KNOWLEDGE GRAPHS FOR DEEPER SENTIMENT UNDERSTANDING IN CODE-MIXED TEXT
Author:	MANYAM THAILE , CHIN-SHIUH SHIEH , DR. M. NAGABHUSHANA RAO , DR. S. K. RAJESH KANNA , DR. RAMARAJU NAGARJUNA KUMAR , DR C MADHUSUDANA RAO , DR. HRUSHIKESH JAIWANT JOSHI , DR. U D PRASAN , RAMARAO NAYAPAMU
Abstract:	KnowSentMix is a hybrid sentiment-reasoning framework for code-mixed social text that selectively combines multilingual contextual representations with compact commonsense knowledge. The method uses an agreement-gated alignment mixture that (i) retrieves and encodes concise knowledge snippets, (ii) grounds concepts to tokens through optimal-transport–based soft alignment, and (iii) fuses text-only and knowledge-only predictions via a gate driven by uncertainty and inter-expert agreement. The system returns a sentiment label together with compact explanatory evidence—top concepts and, when available, trigger spans—improving interpretability with modest computational overhead. Objectives, scope, mathematical formulation, and algorithmic design are accompanied by a full-mode implementation using XLM-R with ConceptNet retrieval. Comparative efficiency measurements and ablation studies indicate that adaptive knowledge gating delivers stable gains on idiomatic, implicit, and noisy code-mixed inputs while preserving efficiency.
Keywords:	Code-Mixed NLP, Sentiment Analysis, Commonsense Knowledge, Conceptnet, COMET, XLM-R, Optimal Transport, Interpretability, Knowledge-Infused Learning, Multilingual Transformers.
Source:	Journal of Theoretical and Applied Information Technology 31^st December 2025 -- Vol. 103. No. 24-- 2025
Full Text