		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.23

Title:	SECURE CLOUD DATA STORAGE USING CONTEXTUAL ENTROPIC CIPHERTEXT-POLICY ATTRIBUTE-BASED ENCRYPTION
Author:	NAGA SWETHA DHULIPUDI, BALAJI SAVADAM
Abstract:	Cloud computing delivers a centralized platform for data storage and commercial applications. To ensure security, the cloud system effectively manages all connected devices, applications, and data. Various existing algorithms fail to optimize the operational delay during data encryption because of ineffective key management, which affects overall security. To overcome this issue, this research proposes Contextual Entropic Ciphertext-Policy Attribute-Based Encryption (CE-CPABE) for secure cloud data storage. CE-CPABE provides improved security and flexibility for cloud data storage by applying attribute-based access control and contextual validation. Integrating entropy-based key generation ensures that every encryption and decryption process is unique and resistant to key reuse, thereby securing data access in dynamic and distributed environments. This approach enhances the integrity, confidentiality, and access efficiency of the sensitive information stored in the cloud. CE-CPABE optimizes cryptographic operations while reducing computational overhead and resource utilization in cloud environments. The experimental results show that CE-CPABE achieves an encryption time of 2.28s and a processing time of 0.62s for a data size of 30MB, which is better than that of existing methods.
Keywords:	Attribute-Based Encryption, Attribute-Based Access Control, Contextual Entropic, Cloud Computing, Secure Cloud Data Storage
Source:	Journal of Theoretical and Applied Information Technology 15^th December 2025 -- Vol. 103. No. 23-- 2025
Full Text

Title:	PUBLIC-PRIVATE PARTNERSHIP IN THE FIELD OF INFORMATION SECURITY AS A TOOL FOR STRENGTHENING NATIONAL SECURITY
Author:	ANATOLII ZINENKO, NATALIIA KRASNOSTANOVA, OLGA LUGACH, INNA KULCHII, OLEKSANDR ORLOVSKYI
Abstract:	Public-private partnerships (PPPs) are successfully used in international practice to strengthen national security, in particular, in the defence sector, which makes the development of this method of public-private interaction relevant in Ukraine. The aim of the study was to assess the impact of PPPs on socio-economic and security aspects in low- and middle-income countries and integrate the experience of leading countries. The research employed the methods of statistical, correlation, and regression analysis. The results of the study showed a significant positive impact of PPP investments on the military strength, logistics, and fire safety of low- and middle-income countries. However, the research did not reveal any significant impact of such investments on ensuring cybersecurity, which requires increased attention from governments to this area. The analysis of the experience of leading countries identified key positive examples of the use of PPPs in the defence and cybersecurity sectors. Key obstacles to the development of PPPs in the cybersecurity sector as an important component of information security were revealed. Priority areas for the development of PPPs in the field of cybersecurity were outlined. The results of the study gave grounds to provide recommendations for Ukraine, which can be used in practice for improving the legislative framework and developing strategies to strengthen national security. The prospects of further research may be the development of proposals for the development of PPPs in the defence sector of Ukraine through the use of specialized platforms for international coordination and information exchange.
Keywords:	Public-Private Partnership, National Security, Information Security, Cybersecurity, Infrastructure, Logistics, Sustainable Development.
Source:	Journal of Theoretical and Applied Information Technology 15^th December 2025 -- Vol. 103. No. 23-- 2025
Full Text

Title:	THE IMPACT OF INDONESIAN FOOD INFLUENCERS ON SOCIAL MEDIA IN SHAPING INDONESIAN CONSUMERS PURCHASE INTENTION FOR FOOD CHOICES DURING OVERSEAS TRAVEL
Author:	KARINA HANANINGSIH, SHEILA GRACIA, SITI PATIMAH, JERRY S. JUSTIANTO
Abstract:	As social media increasingly becomes the primary reference source for culinary information, the role of food influencers has grown more significant in shaping consumer perceptions, preferences, and influencing purchasing decisions—especially in the context of international tourism. This study aims to analyze the impact of Indonesian food influencers on social media in shaping Indonesian consumers' purchase intentions for food choices, particularly when traveling abroad. This research adopts a quantitative approach, utilizing SEM-PLS as the analytical tool to examine the proposed research model. Out of 241 respondents who participated in a mobile-based survey using Google Forms, 225 respondents met the criteria of having been exposed to foreign food review content uploaded by Indonesian food influencers on social media. The findings indicate that influencer expertise and perceived similarity have a significant positive effect on both consumer trust and purchase intention. Interactivity was found to positively influence purchase intention but had no significant effect on trust. Trust plays a significant mediating role in the relationship between expertise and perceived similarity with purchase intention, but not in the relationship between interactivity and purchase intention. These findings highlight the importance of expertise and perceived similarity in building trust toward influencers, as well as the critical role of trust in driving consumers' purchase intentions in the context of international travel. This study contributes new insights into cross-cultural consumer behavior by examining the interplay between same-nationality influencers and consumers within a foreign context, which has been largely overlooked in prior studies. The practical implications of this study can be utilized by stakeholders in the tourism and culinary industries to design promotional strategies based on collaborations with credible food influencers and making them more effective for Indonesian consumers traveling abroad.
Keywords:	Food Influencer, Overseas Travel, Trust, Purchase Intention, Social Media Marketing
Source:	Journal of Theoretical and Applied Information Technology 15^th December 2025 -- Vol. 103. No. 23-- 2025
Full Text

Title:	EPILEPTIC SEIZURE DETECTION FROM EEG SIGNALS USING SVM WITH TIME–FREQUENCY FEATURE EXTRACTION METHODS
Author:	HINDARTO HINDARTO, ADE EVIYANTI, AHMAD AHFAS, EGHA ARYA AFFANDI
Abstract:	Epilepsy is a neurological disorder characterized by recurrent seizures, affecting approximately 50 million people worldwide. Early detection of seizures is crucial for accurate diagnosis and effective treatment. This study proposes several methods for EEG signal feature extraction in the time-frequency domain, including STFT, CWT, DWT, and HHT. The UCI dataset is the data used in this study, this data in 1 second contains 11,500 segments. This dataset is processed through several stages, first the pre-processing stage, then feature extraction using the time-frequency domain. The final stage of classification uses Support Vector Machines (SVM). The classification process is divided into 2 stages, namely the training and testing process, in this case 80% training and 20% testing. Evaluation of the classification results is carried out using evaluation metrics namely recall, F1 score, accuracy, precision, MCC, and Kappa. The results show that the performance reaches 96% more for all methods, except for feature extraction using DWT with 5 features. All methods achieve performance above 96%, except for the feature extraction method using DWT with 5 features. Feature extraction using the STFT method using 14 features and CWT using 49 features produces the highest accuracy of 98.39%, surpassing previous studies that used the SVM method for classification. The worst performance with a value of 86.7% using DWT feature extraction with 5 features. Compared with previous studies, the proposed method provides better accuracy. The study concluded that with a combination of appropriate pre-processing and a suitable feature extraction algorithm, accurate and reliable detection of epileptic seizures can be achieved.
Keywords:	Epilepsy, EEG, Seizure Detection, Time Frequency Domain, SVM.
Source:	Journal of Theoretical and Applied Information Technology 15^th December 2025 -- Vol. 103. No. 23-- 2025
Full Text

Title:	AUDIBLOC: A POST-QUANTUM AUDIO SECURITY SYSTEM WITH FORENSIC WATERMARKING CHAOTIC ENCRYPTION AND BLOCKCHAIN VERIFICATION
Author:	GAYATRI SRUTHI BELLAPUKONDA, NARESH SAMMETA
Abstract:	Audio transmission of information is vital in telemedicine, financial processes, the corporate world, government intelligence, and so on. Nevertheless, the current solutions are being challenged like never before the classic RSA/AES encryption can be broken with quantum computers, traditional watermarking is non-forensics traceable, which is problematic in terms of legal compliance, and centralized verification is prone to single points of failure that can be targeted. This is the case with the AUDIBLOC, which targets these limitations with a four-layer architecture revolution. Our quantum random number generation (QRNG) unlike conventional pseudo-random key generation, true entropy cannot be attained using classical means. Traditional chaotic encryption has the problem of periodicity, but we already have exponentially more confusion and diffusion with our butterfly effect-based implementation. Our quantum forensic watermarking results in 95% successful detection of watermarks, even after compression and filtering in wav extension. This compares favorably to the current watermarking, which is destroyed by even the simplest audio processing. Most importantly, our verification incorporating a blockchain makes single points of failure impossible and allows an immutable audit trial. It also shows better practical performance by experimental validation than other algorithms with encryption speed of 45.2 MB/sec in the range of AES-256 (68.5 MB/sec) but offering quantum resistance, watermark survival that survives 89.2% of signal processing attacks and versus 60-70% of modern schemes, resists both Shor and Grover quantum algorithms. Security analysis demonstrates a 2 256 bits effective key space in the case of audio quality scores (PESQ: 4.32), similar to unencrypted data. AUDIBLOC is a feasible and scalable means of providing post-quantum secure audio, a key factor in ensuring that organizations can secure audio communication with confidence in the authenticity and confidentiality of messages, even as traditional cryptographic underpinnings become no longer useful.
Keywords:	Quantum Cryptography, Audio Encryption, Chaos Theory, Perceptual Evaluation of Speech Quality, Block-chain, Digital Forensics, Post-Quantum Security
Source:	Journal of Theoretical and Applied Information Technology 15^th December 2025 -- Vol. 103. No. 23-- 2025
Full Text

Title:	FACEDEPAI: A DEEP LEARNING FRAMEWORK FOR DEPRESSION DETECTION FROM FACIAL EXPRESSIONS USING FACEDEPNET
Author:	MD ZAINUDDIN NAVEED, DR. SHIVAMPETA APARNA
Abstract:	Depression is a widespread mental health problem with a significant effect on productivity and quality of life. Early, objective detection is still a crucial problem as existing diagnostic methods like clinical interviews and self-reported questionnaires are subjective and resource demanding. Advances in artificial intelligence have facilitated the construction of an automatic depression detection system based on multiple modal data signals, such as text, speech, and image. However, the current image-based methods may overfit to small datasets and not focus on depression-specific characteristics with poor interpretability, leading to their limited effectiveness in clinical applications. To address the gaps above, we propose presenting FaceDepAI, a deep learning system designed for binary depression detection from facial expressions, based on our model, FaceDepNet. The architecture combines convolutional neural networks with squeeze-and-excitation attention modules to dynamically reweight feature channels, emphasizing subtle yet clinically significant facial cues. We use data augmentation and dropout regularization to improve generalization, while incorporating Grad-CAM for increasing interpretability by highlighting depression-related regions on the face. The framework was tested on the DAIC-WOZ dataset with five-fold cross-validation. The experiment results showed that our proposed method achieved 96.8% accuracy, 96.5% precision, and a recall of 97%, with an F1-score as high as 98.2 % ROC-AUC, which consistently outperformed the baseline models, including ResNet-18, VGG-Face, and EfficientNet without an attention mechanism. These results demonstrate FaceDepAI is a robust, explainable, and clinically meaningful tool for depression detection. The scale can be used for broader, larger-scale mental health screening and as an aid to clinical decision making.
Keywords:	Depression Detection, Facial Expressions, Deep Learning, Attention Mechanism, Explainable
Source:	Journal of Theoretical and Applied Information Technology 15^th December 2025 -- Vol. 103. No. 23-- 2025
Full Text

Title:	REPORTHINK.AI ROLE TO REDUCES SUSTAINABILITY REPORT INFORMATION ASYMMETRY
Author:	SUCI NASEHATI SUNANINGSIH, MUMPUNI WAHYUDIARTI SITORESMI, SUAMANDA IKA NOVICHASARI, WILDAN YUDHANTO
Abstract:	This research aims to examine the effect of Reporthink.AI on information asymmetry. This research uses 120 companies that are listed on the Indonesian Stock Exchange ESG Leader Index. Implementation of Reporthink.AI is measured by a dummy variable. Information asymmetry is measured by bid-ask spread. Data analysis uses fixed-effect regression. Based on data analysis, Reporthink.AI brings significant reduction in information asymmetry, validating its value as a technological advancement for enhanced transparency and credibility for company disclosures. This research has some contributions. First, this research contributes to the literature, especially contributes to extend the signalling theory in the context of AI and sustainability reporting. Second, this research gives new evidence of Reporthing.AI on information asymmetry, especially in the Indonesian Stock Exchange.
Keywords:	ESG Leader Index, Information Asymmetry, Machine Learning, Reporthink.AI, Sustainability Report
Source:	Journal of Theoretical and Applied Information Technology 15^th December 2025 -- Vol. 103. No. 23-- 2025
Full Text

Title:	TEXT2CHAIN: A MODULAR NLP-ENHANCED ORACLE ARCHITECTURE WITH DECENTRALIZED VALIDATION AND HUMAN-IN-THE-LOOP FOR SECURE OFF-CHAIN/ON-CHAIN INTEGRATION
Author:	MERIEM KERMANI
Abstract:	The integration of unstructured textual data into blockchain systems presents a significant challenge, primarily due to the semantic ambiguity of natural language and the inherent trust issues associated with centralized oracles. This paper proposes a novel, modular oracle architecture enhanced with Natural Language Processing (NLP) to enable a secure, context-aware translation of human-readable text into reliable on-chain smart contract execution. Departing from conventional single-point oracles, our framework employs a decentralized oracle network designed to perform multi-source semantic validation, cryptographic securing, and consensus-based aggregation of NLP outputs prior to blockchain commitment. The architecture operates in three distinct stages: (1) off-chain NLP processing to extract structured data and confidence scores from text; (2) a network of AI-powered oracles that independently validate, enrich, and cryptographically sign results, with a consensus mechanism (e.g., 2-out-of-3) determining the final authenticated payload; and (3) on-chain execution that is triggered only after successful cryptographic verification of the validated data. This design eliminates single points of failure, enhances resilience against data manipulation, and ensures a trust-minimized semantic bridge between off-chain interpretation and deterministic on-chain logic. A proof-of-concept implementation, utilizing spaCy for NLP, Web3.py, an Ethereum testnet, and ECDSA signatures, demonstrates the framework's feasibility in processing domains such as clinical narratives or service reports into auditable blockchain transactions. The evaluation assesses critical performance metrics, including gas cost and latency, and analyzes the security trade-offs between automation, decentralization, and operational efficiency. The primary contribution of this work is a reproducible, modular framework that advances secure blockchain interoperability by effectively bridging unstructured human language with deterministic machine execution for reliable automation across sectors like healthcare, finance, and public administration.
Keywords:	Decentralized Oracle Network, Natural Language Processing (NLP), Semantic Validation, Smart Contracts, Off-chain/On-chain Integration.
Source:	Journal of Theoretical and Applied Information Technology 15^th December 2025 -- Vol. 103. No. 23-- 2025
Full Text

Title:	HYBRID ATTENTION-DRIVEN CNN-GRU FRAMEWORK FOR ROBUST DETECTION OF MULTI-SCALE POWER QUALITY DISTURBANCES IN SMART MICROGRIDS
Author:	M.DEVIKA RANI, V.SAI GEETHA LAKSHMI, M.V.RAMESH, 4KARUNAKAR KANCHETI, PREMA KANDASAMY, P. MUTHU KUMAR
Abstract:	Smart microgrids are becoming increasingly interconnected with renewable energy sources and nonlinear loads, which means that power quality (PQ) disturbances such voltage sags/swells, harmonics, transients, and frequency deviations are becoming more complex and dynamic. It is fairly uncommon for conventional rule-based or signal processing approaches to miss the multi-scale features and spatiotemporal features of such disruptions. In order to properly identify and categorize power quality events, this research introduces a novel deep learning architecture for smart microgrids that integrates a Convolutional Neural Network (CNN) with a Gated Recurrent Unit (GRU). The GRU keeps track of the event's temporal relationships, while the CNN part uses spatial and frequency-domain data from current and voltage waveforms to extract features. Experimental results demonstrate significant improvements in classification accuracy, F1-score, and inference time when compared to baseline models such vanilla CNN, LSTM, and classic SVM-based classifiers.
Keywords:	Deep Learning-Based Classification; Gated Recurrent Unit (GRU); Power Quality Disturbances; Smart Microgrids; Spatiotemporal Feature Extraction
Source:	Journal of Theoretical and Applied Information Technology 15^th December 2025 -- Vol. 103. No. 23-- 2025
Full Text

Title:	AUTOMATED CLASSIFICATION AND COUNTING OF VEHICLES USING DEEP LEARNING APPROACH
Author:	AMIR RAZA, JOHARI ABDULLAH, REHMAN ULLAH KHAN, IRWANDI HIPNI MOHD HIPINY, HAMIMAH UJIR
Abstract:	The rapid increase in automobile traffic in and around cities presents significant challenges for transportation management, infrastructure planning, and overall system viability. Conventional vehicle classification and counting techniques, which are mostly manual or sensor-based, are often constrained by inefficiency, labor intensity, and limited scalability in real-time applications. This paper proposes a deep learning approach for real-time classification and enumeration of vehicles using one-dimensional signals from piezoelectric sensors. The 1D-CNN was trained, using convolutional layers with small kernel sizes to stack, so that the temporal dependencies of sensor signals could be well learned, pooling and fully connected layers used to extract features with high strength. The model was trained using a hybrid dataset with field-collected sensor data and synthetically generated signals generated with traffic simulation tools to cover a classes of vehicles and road conditions, which guarantees the ability to scale and generalization. The proposed model is shown to perform well as it has a classification accuracy of 0.99, and mean Average Precision (mAP) of 0.98 on five different vehicle classes. Moreover, its performance was checked with the unseen data, which proved high generalization ability. The solution has potential to be deployed on edge computing devices because it is computationally efficient to support a realistic application of traffic monitoring.
Keywords:	Vehicle Classification, Vehicle Counting, 1D Convolutional Neural Network (1D-CNN), Road Energy-Harvesting Sensors, Intelligent Transportation System (ITS)
Source:	Journal of Theoretical and Applied Information Technology 15^th December 2025 -- Vol. 103. No. 23-- 2025
Full Text

Title:	SUPERPIXEL BASED CLASSIFICATION FOR GRAPH NEURAL NETWORK(GNN) IN LEUKEMIA IMAGES
Author:	B. REVATHI, M. KALIAPPAN, P. CHANTHIYA, S.V. ANANDHI, CALLINS CHRISTIYANA.C, S.K. KEZIAL ELIZABETH
Abstract:	Patient survival depends heavily on accurately and quickly detecting leukemia, but traditional diagnostic methods can be time-consuming and subjective. We present a innovative approach that utilizes Superpixel-based image segmentation and graph neural networks (GNN) to classify leukemia in microscopic cell images. The C-NMC dataset converts each image into a graph, with nodes representing meaningful regions (Superpixels) and edges capturing spatial relations. The accuracy of our model is 99.56%, and it performs well in Precision, Recall, and F1-score metrics. This method not only improves accuracy but also makes it easier to model spatial patterns in medical images.
Keywords:	Leukemia Detection, Graph Neural Networks, Superpixel Segmentation, Graph Attention Network, Image Classification, Medical Imaging
Source:	Journal of Theoretical and Applied Information Technology 15^th December 2025 -- Vol. 103. No. 23-- 2025
Full Text

Title:	AN EARLY BREAST CANCER DETECTION USING EFFICIENT NET BASED TRANSFER LEARNING MODEL
Author:	M V GANESWARA RAO, K PRASUNA, MALLAMPATI PURNA KISHORE, SURESH BABU CHANDOLU, 5VENKATA NARAYANA T, SIVAGANGA BADIPATI, S SINDHURA
Abstract:	Since breast cancer is still one of the top causes of death for women globally, effective and precise early detection techniques are desperately needed. Despite their effectiveness, traditional imaging-based diagnostic techniques like mammography and ultrasound are frequently constrained by noise, mistaken human interpretation, and decreased precision in dense breast tissues. This work investigates the use of transfer learning models for early breast cancer detection in order to address these issues. To distinguish between benign and malignant tumors, transfer learning uses pre-trained convolutional neural networks (CNNs), such as VGG16, ResNet50, InceptionV3, and Efficient Net, that have been refined on publically accessible breast imaging datasets. Reusing deep feature representations from extensive natural picture datasets (like ImageNet) allows the models to maintain excellent accuracy with fewer training samples and less computing power. To improve feature extraction and generalization, the suggested method includes pre-processing stages such picture normalization, data augmentation, and contrast enhancement. Results from experiments show that transfer learning models perform better than traditional machine learning classifiers, attaining higher recall, accuracy, and precision across several modalities. With a detection accuracy of over 96%, EfficientNet-B3 offered the best balance between accuracy and computational efficiency among the models that were tested. Additionally, to improve clinical transparency, Grad-CAM imaging is used to highlight discriminative tumour areas and interpret model predictions. This study demonstrates that, particularly in medical settings with limited data, transfer learning provides a strong and scalable foundation for early breast cancer identification. According to the results, early diagnosis, patient prognosis, and healthcare outcomes could all be considerably enhanced by incorporating transfer learning-based diagnostic tools into clinical workflows. For real-time screening and decision assistance, future work will integrate cloud-based diagnostic systems and fuse multimodal data.
Keywords:	CNN, Breast Tissues, Efficient Net, Cancer.
Source:	Journal of Theoretical and Applied Information Technology 15^th December 2025 -- Vol. 103. No. 23-- 2025
Full Text

Title:	CUDA-ENHANCED YOLO FEATURE EXTRACTION WITH RF–CATBOOST ENSEMBLES FOR HIGH-PERFORMANCE PRODUCT REVIEW SENTIMENT DETECTION
Author:	Dr. VANCHA MAHESHWAR REDDY,SRILAKSHMI KAZA, CH.RAMBABU, Dr.A.SRI NAGESH, V.S.N. MURTHY, M PRAVEEN, SARAVANAN G, B.SRILAKSHMI, Dr. E.SREEDEVI,Dr. SIVA KUMAR PATHURI
Abstract:	Sentiment analysis is essential in retrieving opinions, reviews and written statements to predict emotions based on Natural Language Processing (NLP). It categorizes the text in sentiments like positive, negative or neutral. In case of labeled datasets, the customer feedback can be grouped into scales such good, better, best, or bad, worse, worst, and subsequently converted into sentiment categories. The fast expansion of the World Wide Web has created massive amounts of user-generated content opinions and feelings and reviews that are used in business by e-commerce platforms, such as Amazon, Flipkart, and social platforms, such as Twitter and Facebook. Opinion mining has now taken a very important role in business analytics where customer opinion is directly used in creating and maintaining competitiveness in a product. In this paper, a CUDA-Optimized Sentiment Analysis Framework based on YOLO to extract features and a hybrid RF-CATBoost ensemble to classify the sentiments is proposed. YOLO takes both textual and pictorial aspects, and RF-CATBoost reduces bias and variance as they aim to achieve increased accuracy. The proposed YOLO + RF-CATBoost model is benchmarked on the Amazon mobile review dataset against deep learning baselines such as ResNet-50 and RegNetY and it is more accurate and faster. The framework, which involves the use of GPU-accelerated CUDA processing, has high performance, scalability, and efficient sentiment detection to support e-commerce decision-making.
Keywords:	Sentiment Analysis, YOLO , RF–CATBoost Ensemble, CUDA Parallel Processing, Product Review Classification, Deep Learning (ResNet-50, RegNetY)
Source:	Journal of Theoretical and Applied Information Technology 15^th December 2025 -- Vol. 103. No. 23-- 2025
Full Text

Title:	MULTI LEVEL USER VALIDATION MODEL WITH ASYMMETRIC CRYPTOGRAPHY FOR CLOUD DATA SECURITY
Author:	K.VENKATESWARARAO, P MANIKYA PRASUNA, KALPANA SANJAY PAWASE, T.V.N.PRASANNA, KATAKAM RANGANARAYANA, V.LAKSHMAN NARAYANA
Abstract:	Even though cloud computing makes data storage and service delivery more affordable, adaptable, and scalable, it is still particularly susceptible to data breaches, insider assaults, and unwanted access. In light of these difficulties, the authors of this study present a novel approach to authentication accuracy and data protection in cloud environments: the Multi-Level User Validation Model with Asymmetric Cryptography (MLUVM-AC). To prevent identity spoofing and illegal data modification, the architecture combines public-private key encryption with multi-stage user validation to guarantee that encrypted data may only be accessed by certified users. The suggested approach offers a layered authentication technique with asymmetric key management, which is an improvement over traditional systems that use symmetric encryption and single-layer password validation. This leads to better performance and improved security. If we compare our model to the current state of the art, we find that it generates keys 17% faster, improves data security correctness by 18%, and improves validation efficiency by 20%. This study adds to the existing body of knowledge by developing a stable, efficient, and extensible method for protecting data in ever-changing cloud environments through the integration of multi-level authentication with asymmetric cryptography.
Keywords:	Cloud Computing, Data Security, User Authentication, Authorization, Asymmetric Cryptography.
Source:	Journal of Theoretical and Applied Information Technology 15^th December 2025 -- Vol. 103. No. 23-- 2025
Full Text

Title:	ENHANCED DENGUE FORECASTING WITH ANC-DEFO: A HYBRID FEATURE OPTIMIZATION FRAMEWORK
Author:	P.MUTHUSELVI, DR.S.DHANASEKARAN, DR.V.VASUDEVAN
Abstract:	Dengue fever remains a persistent and growing public health challenge, especially in tropical and subtropical regions where it causes significant morbidity and mortality. The complexity and seasonality of dengue transmission necessitate the development of robust and accurate prediction models for early intervention and resource planning. However, the predictive performance of traditional machine learning models often suffers due to high-dimensional data, irrelevant features, and overfitting. This study introduces ANC-DEFO, an Adaptive Neuro-Classifying model integrated with Differential Evolution and Fuzzy Optimization, to enhance the accuracy and efficiency of dengue outbreak prediction. The proposed framework employs intelligent dynamic feature selection to isolate the most relevant environmental and temporal features, thereby improving model generalization and reducing computational cost. Experimental evaluation was conducted using real-world seasonal dengue data from Tamil Nadu, India. The performance of ANC-DEFO was compared against conventional models such as LSTM, Random Forest, SVM, and Logistic Regression. The results indicate that ANC-DEFO achieved a prediction accuracy of 95.5% and a significantly lower RMSE (2.10) compared to LSTM (accuracy 86.9%, RMSE 4.97), demonstrating its superior capability in handling noisy and nonlinear epidemiological data. The model provides a scalable framework for integration into a system for realtime health surveillance and can be modified for other climate-sensitive or vector-borne diseases. By facilitating early outbreak detection, ANC-DEFO shows potential for improving public health readiness and epidemic response strategic planning.
Keywords:	Dengue Prediction, Machine Learning, Feature Selection, Differential Evolution, ANC-DEFO, Public Health.
Source:	Journal of Theoretical and Applied Information Technology 15^th December 2025 -- Vol. 103. No. 23-- 2025
Full Text

Title:	ENSEMBLE DEEP LEARNING WITH ADAPTIVE FEATURE FUSION FOR CORONARY ARTERY DISEASE PREDICTION FROM ECG SIGNALS
Author:	Dr.B.KALPANA, Dr.D.SARASWATHI, KROVVIDI S B AMBIKA, DR.T.VENGATESH
Abstract:	Coronary Artery Disease (CAD) is a leading cause of mortality worldwide. The 12-lead Electrocardiogram (ECG) is a primary, non-invasive, and cost-effective tool for initial screening. While deep learning has shown promise in automating ECG analysis, existing models often underutilize the complex, multi-lead information and struggle with the subtle and heterogeneous manifestations of CAD. This paper proposes ECG-EnsembleFuseNet, a novel ensemble framework that synergistically combines multiple deep learning architectures via an adaptive feature fusion mechanism for enhanced CAD detection. Our approach leverages an ensemble of specialized Convolutional Neural Networks (CNNs) and a Transformer model to capture both localized morphological anomalies and global contextual dependencies across leads. The core innovation is an adaptive fusion module that learns to dynamically weight and integrate features from each ensemble member and each ECG lead, focusing the model on the most salient predictors. Evaluated on a large dataset of 12-lead ECG records from the PTB-XL database, ECG-EnsembleFuseNet achieved an accuracy of 94.7%, a sensitivity of 93.8%, and an AUC-ROC of 0.981, significantly outperforming standalone ResNet (91.2%, AUC: 0.952), InceptionTime (92.1%, AUC: 0.963), and a standard voting ensemble (93.0%, AUC: 0.972). The model's adaptive attention weights provide a form of intrinsic explainability, highlighting the contributory importance of different leads and temporal segments, which aligns with clinical expertise. This work demonstrates the significant benefit of learned, adaptive fusion over static ensemble methods for robust and interpretable CAD prediction.
Keywords:	Coronary Artery Disease, ECG Analysis, Deep Learning, Ensemble Learning, Feature Fusion, Explainable AI, Medical Signal Processing.
Source:	Journal of Theoretical and Applied Information Technology 15^th December 2025 -- Vol. 103. No. 23-- 2025
Full Text

Title:	AN ANALYTICAL MACHINE LEARNING MODEL FOR PREDICTING DEFECTS IN IoT BASED APPLICATIONS
Author:	MANUJAKSHI B C, SHASHIDHAR T M, LATHARANI T R, RENUKADEVI S, N SHIVAKUMAR
Abstract:	In past half decade, there has been considerable progressive advancement towards performing big data analysis over large scale complex network of Internet-of-Things (IoT) using machine learning approach for performing predictive analysis. The proposed study presents a simplified approaches to perform defect analysis using machine learning in large scale IoT based applications like healthcare services. In this study various distinct sensory data are obtained from multiple sources in healthcare services which are further subjected to data aggregation and preprocessing which not only improves the data quality but also reduces the computational burden of training operation. The proposed system is evaluated with multiple machine learning model like random forest which can perform identification as well as classification of defect with respect to its logical classes defined in this study model. The study outcome shows that random forest offers 45% of improved accuracy and 79% of faster processing in comparison to other frequently used machine learning models.
Keywords:	Internet-of-Things, Machine Learning, Prediction, Accuracy, Aggregation
Source:	Journal of Theoretical and Applied Information Technology 15^th December 2025 -- Vol. 103. No. 23-- 2025
Full Text

Title:	A SYSTEMATIC REVIEW OF PERSISTENT AI INTERFACES IN EMBEDDED SYSTEMS: BRIDGING TECHNICAL PERFORMANCE AND USER EXPERIENCE
Author:	ALAN ISAAC TRINIDAD GONZÁLEZ, ELENA FABIOLA RUIZ LEDESMA
Abstract:	This systematic review examines the impact of persistent artificial intelligence interfaces in embedded devices on the usability and user experience of adults. We start by evaluating the performance of these systems and then shift focus to subjective experiences, considering factors such as latency, autonomy, and functionality. Following PRISMA guidelines, we searched Scopus and Springer Nature Link for studies published between 2020 and 2025, focusing on a PICO-formulated question about interactions with these “always-ready” embedded devices. Only studies involving local or hybrid computing that reported on user experience or perceived performance were included. The findings consistently show that persistence is achieved through computational proximity, utilizing voice, vision, or other sensory processes for tasks like word recognition, monitoring in public spaces, and intelligent battery management. As part of our contribution, we outline several implications aimed at establishing quality criteria for embedded devices, useful for both design and evaluation. This framework extends beyond mere computational metrics; it provides guidance for enhancing everyday user experiences, addressing operational goals such as maintaining perceptible thresholds of naturalness, ensuring continuity without delays or lags, and preventing overheating during prolonged use. Overall, it proposes methods to “measure what truly matters” without requiring oversized architectures, while incorporating strategies that make persistence perceptible without intruding on or compromising usability.
Keywords:	Artificial intelligence, Embedded devices, Latency, Pervasive interfaces, User experience.
Source:	Journal of Theoretical and Applied Information Technology 15^th December 2025 -- Vol. 103. No. 23-- 2025
Full Text

Title:	DIGITAL TRANSFORMATION OF LOGISTICS: THE ROLE OF IT IN IMPROVING SUPPLY CHAIN EFFICIENCY
Author:	HASSAN ALI AL-ABABNEH , IBRAHIM MAHMOUD SIAM
Abstract:	Logistics digital transformation is a fundamental accelerator of both efficiency and resilience for supply chains today. Rapidly growing complexity in international trade and rising consumer demands are driving the adoption of advanced information technologies (IT) to increase operational flexibility and decision-making capabilities. The traditional logistics system has developed to a certain degree; however, it also has some disadvantages, such as poor integration, low flexibility, and poor real-time ability to respond. This paper aims at investigating the role of IT (Big Data analytics as a specific IT) on the betterment of supply chain performance. Using a broad methodological framework to evaluate IT impacts on KPIs of different logistics process, the study relies on a Big Data analytics framework. The results revealed that the application of big (demand constraint) on Big Data analytics leads to savings in operational cost up to 20% and more improvement on the average orders delayed and demand forecasting by more than 30% and more than 25% respectively. Novelty The contribution or originality of this work is on that the integrated model has been developed both Big Data analytics and supply chain management theories into problems, this is a holistic vision to digital transformation. The practical implications of the research are fairly straightforward as it offers non-expert logistics managers pragmatic advice as to IT's application in terms of enhancing the supply chain's efficiency and competitive power. The results of this research will provide to deepen the theoretical framework of the IT in the logistics field, and would offer a way to apply IT to real cases.
Keywords:	Information Technology, Supply Chain Efficiency
Source:	Journal of Theoretical and Applied Information Technology 15^th December 2025 -- Vol. 103. No. 23-- 2025
Full Text

Title:	ADVERSARIALLY RESILIENT REMOTE SENSING IMAGE CLASSIFICATION THROUGH CROSS-SPECTRAL ATTENTION-FUSED TRANSFORMER MODELING (CSAF-VIT)
Author:	HEMASHREE P, N VALLIAMMAL
Abstract:	Adversarial threats pose significant challenges to the reliability of remote sensing image classification, particularly in defense and surveillance applications. The proposed CSAF-ViT framework introduces a cross-spectral attention-fused vision transformer designed to enhance resilience against pixel-level perturbations and preserve spatial-spectral integrity. The model begins by decomposing multi-band satellite imagery, followed by band-specific feature encoding and attention-guided fusion to capture intricate cross-band relationships. Transformer-based spatial encoding then enriches global contextual awareness, ensuring consistent representation of spectral semantics. A classification token mechanism is employed for final decision mapping, while adversarial training fortifies the model against gradient-induced distortions. The integration of spectral coherence and spatial dependency within a unified architecture enables stable feature propagation and decision reliability under manipulated inputs. CSAF-ViT offers a structured defense pipeline that emphasizes spectral alignment, attention consistency, and robust learning dynamics, establishing a reliable foundation for secure and accurate remote sensing classification in critical operational scenarios.
Keywords:	Remote Sensing Classification, Adversarial Robustness, Cross-Spectral Attention, Vision Transformer, Spectral Feature Fusion, Secure Image Recognition
Source:	Journal of Theoretical and Applied Information Technology 15^th December 2025 -- Vol. 103. No. 23-- 2025
Full Text

Title:	DAFDN: A DUAL ATTENTION NETWORK FOR ROBUST AND INTERPRETABLE FACE FORGERY DETECTION IN VIDEOS
Author:	TENALI ANUSHA , DR. A. SRI NAGESH
Abstract:	The rapid advancement of deep generative technologies has enabled the creation of highly realistic face-manipulated videos, commonly referred to as deepfakes. While these techniques offer benefits in media and entertainment, their misuse poses significant threats to privacy, trust, and information security. This paper introduces a novel deepfake detection framework called the Dual Attention Forgery Detection Net-work (DAFDN), which integrates two specialized attention modules: The Spatial Reduction Attention Block (SRAB) and the Forgery Feature Attention Module (FFAM). These modules are designed to enhance the model's ability to capture subtle artifacts left by tampered facial regions. Built upon the EfficientNet-B4 backbone, DAFDN is evaluated on two benchmark datasets—FaceForensics++ and Deepfake Detection Challenge (DFDC). Experimental results demonstrate that DAFDN achieves strong performance, with AUC scores of 0.945 and 0.911 on FF++ and DFDC respectively, outperforming several state-of-the-art meth-ods. The proposed model not only improves detection accuracy but also offers interpretability through attention-based heatmaps, making it a robust tool for combating video-based face forgery.
Keywords:	Deep Fake Detection, Spatial Attention, Channel Attention, Efficientnet-B4, Dual Attention Network, Video Forensics, Forgery Localization, Manipulated Media Analysis.
Source:	Journal of Theoretical and Applied Information Technology 15^th December 2025 -- Vol. 103. No. 23-- 2025
Full Text

Title:	VIABLE SENSOR DATA AGGREGATION AND COMMUNICATION SCHEME FOR AIDING DIAGNOSIS PRECISION OF INTERNET OF THINGS-BASED HEALTH MONITORING SYSTEMS
Author:	Dr. L. MOHANA KANNAN, K. DEEPIKA, S. RAMYA, P. VIGNESHKUMAR, M.VALLIKKANNU, Dr. R. KANNAN
Abstract:	The rapid growth of the Internet of Things (IoT) has greatly improved smart healthcare systems with wearable sensors and smart data processing. However, continuous health monitoring often encounters problems like data loss, delays in transmission, and communication errors. These issues can lower the accuracy of disease diagnosis. To tackle these challenges, this article suggests a Viable Data Aggregation Scheme (VDAS). This scheme aims to boost diagnosis accuracy by reducing data loss and improving communication reliability in IoT healthcare settings. The proposed scheme uses a distributed federated learning (FL) model to check data availability and identify missing or delayed information during ongoing monitoring periods. By matching sensed and received data over consecutive intervals, VDAS helps consolidate communications from wearable sensors and reduce transmission errors. Experimental results show that this method increases data availability, aggregation efficiency, and communication rate by 9.19%, 8.17%, and 10.31% respectively across various sensing intervals. Although the system has synchronization issues due to non-uniform sensing times, future work will focus on creating interval-based responsive communication methods. This will aim for more reliable and prioritized data delivery in asynchronous healthcare monitoring environments.
Keywords:	Data Error, Diagnosis Accuracy, FL, Healthcare Monitoring, IoT
Source:	Journal of Theoretical and Applied Information Technology 15^th December 2025 -- Vol. 103. No. 23-- 2025
Full Text

Title:	ASPECT-BASED SENTIMENT ANALYSIS USING GLOBAL HYPERBOLIC TANGENT SIGMOID DEEP NEURAL NETWORK WITH FENNEC FOX OPTIMIZATION
Author:	SREEMATHY JAYAPRAKASH , PRASATH NITHIYANANTHAM
Abstract:	Analyzing human speech sentiments is a natural process for humans but an extremely challenging one for machines because of the difficulty in interpreting underlying emotions from beneath content-based meaning. While content-based data analyzes well using traditional sentiment analysis models, it fails to decipher contextual emotions buried deep within speech. Earlier research mainly addressed basic sentiment tagging without considering fine-grained contextual and affective features, leading to poor accuracy and generalizability across various emotional classes. To overcome these limitations, this study proposes a novel approach for sentiment analysis from signal data with the IEMOCAP dataset: Global Hyperbolic Tangent Sigmoid Deep Neural Network with Fennec Fox Optimization (GHTSDNNet-FFO). The approach starts with pre-processing by Deep Attentional Guided Image Filtering (DAGIF), then feature extraction using Style-Based GAN Encoder (SB-GAN Encoder) that retains subtle emotional features. GHTSDNNet is applied for classification, which is a hybrid model combining Global Attention Network (GLOBATTNET) and Hyperbolic Tangent Sigmoid Deep Neural Network (HTSDNN) and additionally optimized with Fennec Fox Optimization (FFO) to enhance learning efficiency. Experimental outcomes on IEMOCAP dataset prove outstanding performance with 99.85% accuracy, F1-Score of 99.58% and precision of 99.62%, which proves the strength and efficiency of the introduced model in emotion-aware sentiment analysis.
Keywords:	Deep Attentional Guided Image Filtering, Fennec Fox Optimization, Global Attention Network, Hyperbolic Tangent Sigmoid Deep Neural Network, Style-Based GlobAttNet Encoder.
Source:	Journal of Theoretical and Applied Information Technology 15^th December 2025 -- Vol. 103. No. 23-- 2025
Full Text

Title:	TECHNOLOGY OF BLOCKCHAIN SYSTEMS: NEW OPPORTUNITIES AND ISSUES IN DATA SECURITY AND RETENTION
Author:	SERGII BATAIEV , VIKTOR KYRYCHENKO , VIKTOR OSTAPCHUK , VIKTOR KRASNOSHCHOK , SERHII MARTYNIUK
Abstract:	In the modern digital environment, there is a growing need for reliable mechanisms for data protection and storage, which is due to the risks of cybercrime, information leaks and manipulation of digital records. Blockchain technologies offer decentralized solutions that can provide increased security, transparency and immutability of data, which makes them promising for use in the financial sector, medicine, e-government and logistics. The purpose of this study is to analyze the opportunities and challenges of blockchain in the field of security and information storage. The work uses methods of comparative analysis, empirical threat modeling and expert assessment of blockchain implementation in various industries. The study confirmed that blockchain is able to significantly reduce the risks of unauthorized access to data and manipulation of digital records, thanks to the use of cryptographic protection methods and consensus algorithms. However, it was found that existing blockchain networks face scalability problems, high energy consumption and the lack of a unified regulatory framework, which limits their implementation. Possible solutions are proposed, including the transition to more energy-efficient algorithms, such as Proof-of-Stake, the development of quantum-resistant cryptographic methods, and the harmonization of regulatory approaches. The results obtained have practical significance for the financial, government, and technology sectors, contributing to the implementation of secure digital solutions based on blockchain. Further research should be aimed at improving scaling technologies and increasing the security of blockchain systems in the context of the development of quantum computing.
Keywords:	Blockchain, Data Security, Cryptography, Consensus Algorithm, Scalability, Financial Technology, E-Government
Source:	Journal of Theoretical and Applied Information Technology 15^th December 2025 -- Vol. 103. No. 23-- 2025
Full Text

Title:	GAME-THEORETIC SELF-SUPERVISED STRATEGIES FOR INTELLIGENT ANOMALY DETECTION IN DYNAMIC SURVEILLANCE ENVIRONMENTS
Author:	Dr. VUDA SREENIVASA RAO, Prof. CHIN-SHIUH SHIEH, Prof. SIVA SHANKAR S, Prof. PRASUN CHAKRABARTI
Abstract:	Detecting anomalies in large-scale video surveillance is challenging due to diverse scene variations and the limited availability of labeled anomalous data. Whereas existing models employed supervised deep networks or rule-based heuristics, this proposed GTSSL framework presents a game-theoretic self-supervised paradigm that enhances system adaptability without requiring large annotated datasets. The method redesigns anomaly detection in IT-driven surveillance by combining decision theory with autonomous learning, a union that is seldom attempted in existing work. The unique contribution of this research is the development of the Game-Theoretic Self-Supervised Learning (GTSSL) framework, which bridges the gap between self-supervised feature learning and adaptive decision-making in anomaly detection. In contrast to traditional deep learning approaches that are based on labeled data or fixed optimization, this framework combines game-theoretic strategy formation, GAN-based augmentation, and CNN-LSTM hybrid modeling, thus presenting a novel paradigm in intelligent surveillance analytics. This methodology puts forward the overall corpus of knowledge in IT by codifying anomaly detection into a strategic optimization problem, enhancing flexibility, robustness, and online learning in complex monitoring situations. The added knowledge, therefore it forms an enlarged step over previous incremental progress in that it brings a theoretically sound, practically scalable, and computationally feasible model translatable to various fields of security and surveillance. The framework is implemented using Python with TensorFlow and OpenCV libraries and evaluated on the Kaggle CCTV Activity Identification Collection and approximately 200 clips per class. Experimental analysis demonstrates superior performance of GTSSL, achieving 98.3% accuracy, 97.5% precision, 97.4% recall, and an F1-score of 97.0%, surpassing CNN (72%), SVM (72.1%), and CNN-RNN (84%) models by a margin of 6–9%. Results confirm that GTSSL enhances resilience against distributional shifts and adversarial patterns while maintaining computational efficiency for real-time deployment. These findings validate the efficacy of combining self-supervised representation learning, adversarial augmentation, and game-theoretic decision-making in anomaly detection systems for dynamic surveillance environments.
Keywords:	Anomaly Detection, Game-Theoretic Learning, Generative Adversarial Networks, Convolutional Neural Networks, Long Short-Term Memory, Video Surveillance
Source:	Journal of Theoretical and Applied Information Technology 15^th December 2025 -- Vol. 103. No. 23-- 2025
Full Text

Title:	SOFTWARE REFACTORING OPPORTUNITIES PREDICTION ON COMMENT-BASED FEATURES USING MACHINE LEARNING CLASSIFIERS
Author:	ARCHANA PATNAIK, NEELAMADHAB PADHY, LOV KUMAR, RASMITA PANIGRAHI
Abstract:	Context: Software Refactoring focuses on internal improvement of the code base without altering its external behavior. For enhancing the quality of real-time open-source projects, it is highly essential to perform complexity analysis followed by code restructuring. Objectives: In this work, we have created a hybrid model by combining different machine learning classifiers to predict refactoring mechanisms on code comments. Quality assessment of a software project is performed on comment tags using machine learning models. Methodology: Initially, the comment-based code snippets are pre-processed to separate the clean and dirty code snippets. The implementation is done by using different classifiers like Naive Bayes, Random Forest, Decision Trees, Logistic Regression, K-Nearest Neighbor, and Support Vector Machine. Result: We have identified six comment tags for refactoring prediction, and TODO comments is high as compared to others. We have implemented six different classifiers, and it is observed that the performance of the Random Forest classifier is high with 95% accuracy, 0.95 precision, 0.94 recall, and 0.93 f-measure. Conclusion: It is observed that the performance of sample 3 is better than other samples, and the Random Forest classifier has having highest accuracy. In our future work, we will predict refactoring on Python code.
Keywords:	Refactoring, Code Comments, Machine Learning Classifiers, Software Quality.
Source:	Journal of Theoretical and Applied Information Technology 15^th December 2025 -- Vol. 103. No. 23-- 2025
Full Text

Title:	INTERFERENCE-AWARE SPECTRUM OPTIMAZATION IN 5G DEVICE-TO-DEVICE USING GENETIC ALGORITHM AND SIMULATED ANNEALING
Author:	MOHD AZRULAZWAN JUSOH MOHD YUSOFF, NOR FADZILAH ABDULLAH, ASMA’ ABU SAMAH, MAHAMOD ISMAIL
Abstract:	The rapid expansion of 5G networks promises improved speed, reduced latency, and massive connectivity, yet dense deployments introduce severe interference and spectrum scarcity. This paper presents a novel hybrid Dynamic Spectrum Management (DSM) framework that integrates Genetic Algorithm (GA) and Simulated Annealing (SA) to enhance spectrum allocation in Device-to-Device (D2D) communication. Unlike static methods, the proposed GA-SA model dynamically optimizes spectrum assignments based on real-time Signal-to-Interference-plus-Noise Ratio (SINR) and interference data. Simulations conducted in a 500-user dense 5G scenario demonstrate significant improvements, with average SINR increasing from 8.2 dB (static) to 14.7 dB and average throughput rising from 5.4 Mbps to 12.5 Mbps. Additionally, the 90th percentile throughput improved from 8 Mbps to 19.3 Mbps. These results validate the GA-SA framework’s effectiveness in reducing interference, boosting capacity, and offering a computationally efficient alternative for real-time 5G spectrum management.
Keywords:	5G Networks, Dynamic Spectrum Management, D2D Communication, Genetic Algorithm, Simulated Annealing, SINR, Throughput.
Source:	Journal of Theoretical and Applied Information Technology 15^th December 2025 -- Vol. 103. No. 23-- 2025
Full Text

Title:	REAL-TIME EMAIL SPOOFING DETECTION USING MACHINE LEARNING AND TIMESTAMP ANOMALY ANALYSIS
Author:	ROOBAL, RAHUL SAXENA, VENUS DILLU
Abstract:	Existing rule-based mechanisms (SPF, DKIM, DMARC) mitigate only a fraction of spoofing attempts and fail against timestamp-manipulated or forwarded mails, revealing a persistent knowledge gap in correlating temporal anomalies with spoofing likelihood. This study addresses that gap by introducing a timestamp-driven anomaly-based machine-learning model (XGBoost) for real-time email fraud detection. Unlike prior content-filter or signature-based systems, the proposed framework fuses authentication records, sender reputation, and delay deviations to create new knowledge on temporal-behavioural indicators of spoofing. Results (R² = 0.92–0.94) confirm superior accuracy and practical deployability. Multiple machine learning models, including Ordinary Least Squares (OLS) Regression, Polynomial Regression, and XGBoost, were tested. Results indicate that XGBoost outperforms traditional models, achieving an R² score of 0.92–0.94, making it highly effective for real-time email fraud detection. The study also highlights the strong correlation between email delay anomalies and spoofing behaviour, with spoofed emails exhibiting significantly longer transmission delays. A flowchart-based implementation is provided, demonstrating real-world deployment feasibility. This contributes new insight into how temporal metadata can be operationalized for enterprise-scale, real-time spoofing prevention. Future work will focus on deploying the model as a cloud-based API and expanding the dataset with real-world email samples for further validation.
Keywords:	Email Spoofing, Machine Learning, XGBoost, Cybersecurity, Timestamp Anomaly Detection
Source:	Journal of Theoretical and Applied Information Technology 15^th December 2025 -- Vol. 103. No. 23-- 2025
Full Text

Title:	DIGITAL TOOLS FOR MONITORING THE ELECTORAL PROCESS IN UKRAINE UNDER MARTIAL LAW
Author:	OLEKSANDR HRYHORIEV, LIUDMYLA PAVLOVA, OLENA KARCHEVSKA, GANNA MALKINA, ALINA VOICHUK
Abstract:	Relevance of the research The relevance of the study is determined by the need to ensure electoral integrity, traceability, and trust in digital voting procedures under martial law and an increased level of cyber threats. Aim of the research The aim of the research is the formalization of a digital voting framework to ensure transparency and electoral integrity under martial law through the integration of legal, technical, and organizational solutions. Research methods The research employed the following methods: structural-functional analysis, comparative law, typology and cluster grouping, technological classification and ranking, cross-validation functional modelling, Unified Modelling Language (UML) modelling. Obtained results The study formalizes a framework for digital will expression relevant to martial law restrictions. Structural-functional analysis and comparative law identified critical threats (security, regulatory, cognitive), while cross-validation modelling identified digital technologies with maximum compensatory potential (blockchain, e-/i-Voting, audit trail). The proposed UML architecture provides multi-level authentication, traceable verification, and civic oversight, which guarantees electoral integrity and regulatory validity in times of crisis. Academic novelty of the research The academic novelty of the study is the stratification of electoral threats under martial law and the formalization of a framework for digital expression of will, which integrates multi-factor authentication, blockchain storage, traceological audit, and citizen oversight, ensuring the stability of procedures, the legitimacy of results, and the autonomy of subjects of will declaration. Prospects for further research Prospects for further research include launching a controlled pilot project to implement a digital will declaration framework in a limited jurisdiction with subsequent validation of its operational stability, regulatory compatibility, and behavioural integrativity. Based on the results of the testing, it is appropriate to develop adaptive optimization modules aimed at increasing cyber resilience, minimizing transactional load, and ensuring institutional scalability.
Keywords:	Digital Suffrage Framework, Blockchain Voting Architecture, Multi-Factor Authentication (MFA), End-to-End Verifiability (E2EV), Smart Contracts, Legal Compliance Modelling, Civic Audit Infrastructure.
Source:	Journal of Theoretical and Applied Information Technology 15^th December 2025 -- Vol. 103. No. 23-- 2025
Full Text

Title:	EVALUATION METRICS OF MACHINE LEARNING OPERATIONS (MLOPS)
Author:	NUR FARAH AFIFAH AHMAD SUKRI, WAN MOHD AMIR FAZAMIN WAN HAMZAH, MOHD KAMIR YUSOF, ISMAHAFEZI ISMAIL, HARMY MOHAMED YUSOFF, AZLIZA YACOB
Abstract:	Machine Learning Operations (MLOps) has emerged to address challenges associated with deploying, integrating, monitoring, and scaling machine learning (ML) models in production environments. However, to effectively evaluate how well MLOps improves ML models, it is important to have clear and standardised evaluation metrics for such measurement. This study introduced thematic analysis to compile and map evaluation metrics from previous research, select the most relevant criteria and then define a comprehensive set of metrics to assess MLOps implementations. The key findings presented in this article focus on eight main themes: data management, automation and pipelines, model performance, resource and time efficiency, deployment and scalability, usability and collaboration, monitoring and observability, as well as compliance and security.
Keywords:	Machine Learning (ML), Machine Learning Operations (MLOps), Machine Learning Evaluation Metrics, MLOps evaluation, Thematic Analysis.
Source:	Journal of Theoretical and Applied Information Technology 15^th December 2025 -- Vol. 103. No. 23-- 2025
Full Text

Title:	AI-POWERED RESUME SCREENING SYSTEM FOR SMART HIRING: LEVERAGING NLP AND LARGE LANGUAGE MODELS FOR EFFICIENT AND FAIR RECRUITMENT
Author:	ANGEL HEPZIBAH R, JEYASANTHI J, DURGA DEVI G, 4MARIAPPAN E, KALIAPPAN M, PREETHY REBECCA P, SENTHI S, RAMNATH M
Abstract:	Manual resume screening is a labour-intensive and error-prone process, further complicated by the growing volume of applications in modern recruitment. While traditional Applicant Tracking Systems (ATS) depend heavily on keyword-based filtering, they often overlook contextually relevant candidates and inadvertently introduce bias. This study proposes a novel AI-powered resume screening framework that combines Sentence-BERT (SBERT) for semantic similarity with Large Language Models (LLMs) for context-aware summarization and bias detection. Unlike existing approaches, our system introduces (i) an enhanced multi-stage candidate-job matching pipeline that integrates semantic embeddings with domain-specific fine-tuning (ii) a bias-mitigation layer that detects and minimizes gender and language-based skew in candidate shortlisting, and (iii) a real-time adaptive HR dashboard with explainable candidate ranking. The system incorporates automated resume parsing, cosine similarity scoring, personalized email notifications, and interactive analytics for recruiters. Experimental results on benchmark recruitment datasets demonstrate a notable improvement in candidate-job matching accuracy (78%) compared to existing ATS methods (avg. 65–70%) along with measurable reductions in screening bias and processing time. The proposed approach is scalable, industry-agnostic and customizable, representing a new step toward fair, efficient, and explainable AI in recruitment.
Keywords:	Resume Screening, Natural Language Processing, Sentence-BERT, Large Language Models, Cosine Similarity, Recruitment Automation.
Source:	Journal of Theoretical and Applied Information Technology 15^th December 2025 -- Vol. 103. No. 23-- 2025
Full Text

Title:	DEVELOPMENT OF AN IOT-ENABLED SMART FARMING APPLICATION WITH INTEGRATED FINANCIAL MANAGEMENT AND COLLABORATIVE KNOWLEDGE SHARING
Author:	ZAMREE CHE-ARON, DARANAT TANSUI, ZAWANEE MAKENG
Abstract:	The integration of the Internet of Things (IoT) in agriculture has led to the development of smart farming solutions that enhance planting management and control. This study presents the design and development of a smart farming application that utilizes IoT technology to optimize crop cultivation. The proposed system incorporates IoT sensors to monitor soil moisture, temperature, and water levels in real time, enabling automated irrigation and lighting control. Moreover, the application allows farmers to track crop growth, manage income and expenses, access agricultural knowledge, and interact with a community forum. A usability evaluation, conducted with IT experts and end-users, demonstrated high levels of satisfaction in terms of user interface design, learnability, efficiency, memorability, correctness, security, and overall functionality. The findings suggest that the developed application provides a practical and efficient tool for farmers, reducing operational costs, conserving resources, and improving crop yields. Furthermore, the financial management and knowledge-sharing functionalities contribute to better decision-making and increased collaboration within the agricultural sector.
Keywords:	Internet of Things, Smart Farming, Mobile Application, Planting Management, Remote Monitoring and Control
Source:	Journal of Theoretical and Applied Information Technology 15^th December 2025 -- Vol. 103. No. 23-- 2025
Full Text

Title:	EXPLORING PERFORMANCE OPTIMIZATION THROUGH ADAPTIVE SHADERS IN UNITY-DRIVEN VIRTUAL REALITY AND MOBILE GAMES
Author:	YULIIA YERMOLAIEVA
Abstract:	The study evaluated the effectiveness of various shader configurations within VR and mobile applications using the Unity engine to achieve consistent performance with high visual fidelity under conditions of variable scene complexity. The problem addressed lies in balancing rendering speed and visual quality in resource-constrained environments, such as mobile GPUs and VR headsets, where fluctuations in scene complexity often lead to frame drops and a degraded user experience. Three rendering alternatives were examined, namely: basic shaders (Base-Shader), optimized with fixed parameters (Static-Opt), and adaptive shaders with dynamic load scaling (Adaptive-Shader). The comparative analysis employed an integral rendering efficiency index (0.87–0.91 for Adaptive-Shader, 0.82–0.85 for Static-Opt, and 0.79–0.81 for Base-Shader), which synthesized the average FPS, frame stability, frame processing time, and the visual quality index. The methodology encompassed normalization of indicators, analysis of variance with repeated measures (Repeated Measures ANOVA), the Friedman test, and post-hoc comparisons, as well as regression analysis, to discern the impact of polygonality and shader type on performance. The findings revealed that Adaptive-Shader consistently achieves over 72 frames per second in VR, even amid high-load scenes, while maintaining a visual quality index of approximately 0.9, with power consumption on mobile GPUs ranging from 7.8 to 8.1 watts. Static-Opt presents a satisfactory equilibrium in simple to moderately complex scenes but exhibits diminished stability in high-complexity scenarios. The Base-Shader remains stable in low-complexity scenarios but shows the most significant decline in FPS and quality under challenging conditions. Statistically significant differences (χ² = 12.48; p = 0.0019) corroborated the superiority of Adaptive-Shader over the other configurations, particularly regarding the ΔFPS-index, where the disparity with Base-Shader approached nearly a twofold advantage. In practical terms, Adaptive-Shader is particularly advantageous for VR applications that demand high smoothness and detail. Static-Opt is suitable for mobile games constrained by limited resources, and Base-Shader is appropriate for rapid prototyping and simple scenes. The scientific novelty is underscored by generalizing the interaction patterns among shader type, scene characteristics, and graphic performance requirements, thereby facilitating a judicious integration of adaptive graphics technologies into VR and mobile gaming. In conclusion, the study demonstrates that adaptive shader technology provides a statistically and practically superior balance between FPS stability, visual fidelity, and energy efficiency, making it the most promising pathway for next-generation VR and mobile rendering. Prospects for further investigation include expanding the metric set to incorporate power consumption and adapting the methodology for multi-platform rendering systems.
Keywords:	Unity, Adaptive Shaders, FPS, Frame Stability, Visual Quality Index, Δperf-Index, VR Games, Mobile Games, Non-Parametric Tests
Source:	Journal of Theoretical and Applied Information Technology 15^th December 2025 -- Vol. 103. No. 23-- 2025
Full Text

Title:	BLOCKCHAIN FOR SECURE AND DECENTRALIZED CLOUD COMPUTING ENHANCING DATA PRIVACY AND INTEGRITY
Author:	ANJANEYULU NELLURU , VALAPALA PRABHAVATHI , APPIKATLA NAGA PRAVALLIKA , DR RAGHAVENDER K V , BATTULA SOWJANYA , B YAMINI SUPRIYA, SATHISH KUMAR SHANMUGAM
Abstract:	This paper introduces a new hybrid blockchain-cloud architecture to improve cloud computing's data security, privacy, and integrity. The main contribution of this research is the introduction of blockchain's decentralized ledger and cryptographic characteristics into cloud computing for secure data storage, access control, and data provenance. The blockchain we use is a permissioned network; the consensus mechanism we have chosen is Practical Byzantine Fault Tolerance (PBFT), and we use homomorphic encryption for secure evaluation. The system's performance was measured using transaction time, response time, scalability, and security metrics. The results indicate that blockchain-based integration comes with latency and throughput overhead compared to classical cloud systems. Still, there is a notable increase in security and privacy, achieving higher Total Security Scores (TSS) and Total Privacy Scores (TPS) than classical cloud systems. The scalability experiments show that the proposed system scales well under high loads, although at the expense of some performance. Such hybrid architecture provides a strong, transparent, and secure solution to the cloud environment that can be suitable for industries with a strong need to protect data, such as healthcare and financial. The results indicate the practicability of blockchain security on clouds, which enormously impacts distributed cloud service data integrity and user trust.
Keywords:	Blockchain, Cloud Computing, Data Privacy, Data Integrity, Hybrid Architecture, Security
Source:	Journal of Theoretical and Applied Information Technology 15^th December 2025 -- Vol. 103. No. 23-- 2025
Full Text

Title:	AN EFFICIENT REINFORCEMENT LEARNING-BASED SUPPLIER SELECTION FRAMEWORK THROUGH RISK PREDICTION PROCEDURES
Author:	INDUMATHY M , BHARATHI N
Abstract:	Supply networks have needed structural modifications to react to both positive and negative occurrences such as Industry 4.0 and natural disasters. Both negative and positive things can disrupt corporate business operations and have an impact on their continuation. Selected suppliers can support the organizations against disruptions. To check uninterrupted materials across flow of the Supply Chain Management (SCM) supplier selection and other distribution must be reorganized in light of the dynamics of disaster Industry 4.0 occurrences. SCM is directly impacted by different decision-making processes like supplier evaluation and selection. Several approaches have been used to evaluate suppliers' performance and choose the top suppliers more effectively, enhancing the overall performance at diverse conditions. Yet, the conventional approaches are not sufficient for managing with large volumes of training datasets within a restricted period, minimizing the convergence speed. Also, it is not capable for generating optimal solution in the validation phase and it requires most efficient performance measures. Therefore, this research is also exploring a new supplier selection strategy incorporating heuristic-driven deep learning. In the initial stage, the industrial data are gathered and forwarded to the data cleaning phase to increase the data quality. Then, the data transformation is conducted to enhance the efficiency of information flow without unnecessary complexity. The next stage is the optimal weighted feature selection, where the weights and optimal features are optimized using a new hybrid optimizer termed Position Updated-Archimedes and Fruit Fly Optimization (PU-AFFO) technique. Finally, the selected weighted features are given to the risk prediction performance, where the outcomes related to the risk is acquired using the Adaptive Transformer Bidirectional Long Short-Term Memory with Weighted Bayesian Learning (ATransLSTM-WBL). Here, the parameters of every network are tuned using the same PU-AFFO technique. Finally, based on the predicted risk, Reinforcement Learning is performed for the final supplier selection. At last, to check the developed system performance, various statistical analyses are utilized and the outcomes given statistically powerful GEP model. Here, the designed framework has attained a lower error rate of 1.01, 0.01, 3.27, 1.12, and 5.88 in terms of MEP, SMAPE, MASE, MAE and RMSE measures, which demonstrate the overall efficiency of the designed technique than the conventional methods.
Keywords:	Risk Prediction-based Supplier Selection Model; Position Updated-Archimedes and Fruit Fly Optimization; Transformer Bidirectional Long Short-Term Memory; Weighted Bayesian Learning; Weighted Feature Selection
Source:	Journal of Theoretical and Applied Information Technology 15^th December 2025 -- Vol. 103. No. 23-- 2025
Full Text