		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
March 2025 | Vol. 103 No.5

Title:	EVALUATING FORMAL METHODS FOR VERIFYING SECURITY PROTOCOLS: A CASE STUDY OF TAMARIN, AVISPA, AND PROVERIF
Author:	ADEL HASSAN , ISAM ISHAQ , JORGE MUNILLA
Abstract:	Verifying security protocols using formal methods is crucial to ensure their robustness against cyber threats. Several verification tools, including Tamarin, AVISPA, and ProVerif, offer different methodologies for protocol analysis. However, a comprehensive comparative analysis of these tools under uniform conditions remains limited. This study systematically evaluates these three tools by assessing their verification mechanisms, supported programming languages, and usability. A standardized testing framework was employed to ensure a consistent comparison, focusing on two widely used security protocols: the Diffie-Hellman Key Exchange Protocol and the Needham-Schroeder Public Key Protocol. The findings highlight distinct strengths and weaknesses in each tool. Tamarin demonstrated superior capability in detecting active attacks such as Man-in-the-Middle (MitM) attacks, while ProVerif was more effective in identifying passive attacks like eavesdropping. AVISPA, on the other hand, provided a broader but less detailed security analysis. These insights help researchers and practitioners select the most appropriate tool based on protocol complexity and security requirements. Unlike prior research that focused on individual tools, this study offers a comprehensive empirical comparison, providing deeper insights into their practical effectiveness and limitations. The results contribute to enhancing security protocol verification methodologies and informing future improvements in formal verification tools.
Keywords:	Formal Methods, Security, Security Protocol, Protocol Modeling, Verification Processes, Testing Tools.
Source:	Journal of Theoretical and Applied Information Technology 15^th March 2025 -- Vol. 103. No. 5-- 2025
Full Text

Title:	INTEGRATING TAM, TPB, AND SCT IN PREDICTING CAATS ADOPTION: A TRIPLE LENS APPROACH
Author:	ATHA ULIMA ARDELIA, CARISSA FELICIA FANGASADHA, RINDANG WIDURI
Abstract:	The swift advancement of technology has created a notable transformation in the auditing field. External auditors are urged to adapt Computer-Assisted Audit Techniques (CAATs) utilization in order to improve their performance. However, CAATs adoption in Indonesia has not yet been mandatory by regulations. Therefore, this research aims to examine the factors influencing auditor’s CAATs adoption by integrating Technology Acceptance Model (TAM), Theory of Planned Behavior (TPB) and Social Cognitive Theory (SCT). Questionnaires were distributed to 100 auditors working at Public Accounting Firm (PAF) in Indonesia. This study employs Structural Equation Modelling (SEM) in SmartPLS to analyze the relationships between variables. Key findings of this study discovered that perceived ease of use (PEOU) has positive influence towards perceived usefulness (PU) of auditors, PU has positive influence towards attitude (AT), AT has positive influence towards behavioral intention (BI), subjective norm (SN) has positive influence towards BI, perceived behavioral control (PBC) has a positive influence towards BI, self-efficacy (SE) has positive influence towards BI, and BI has positive influence towards auditors’ actual behavior in using CAATs. However, the PEOU does not have a positive influence towards AT. In conclusion, the constructs used in this study are linked and capable of predicting auditors’ CAATs adoption. This research could contribute to audit research by supporting TAM, TPB, and SCT, while also developing a model to predict the acceptance and rejection of CAATs adoption. Furthermore, this study could be used by decision makers to foster technology-driven audit process.
Keywords:	CAATs, audit technology, Technology Acceptance Model, Theory of Planned Behavior, Social Cognitive Theory
Source:	Journal of Theoretical and Applied Information Technology 15^th March 2025 -- Vol. 103. No. 5-- 2025
Full Text

Title:	THE MEDIATING EFFECT OF CUSTOMER FEEDBACK ON THE RELATIONSHIP BETWEEN INSTAGRAM CONTENT AND BRAND AWARENESS AMONG MILLENIALS AND GEN Z: A REGRESSION ANALYSIS
Author:	AISYA PUTRI ZANUARIZQI , IRMAWAN RAHYADI , ANIS FAIQOH NURULLITA , DIVIA INDIRA ARIFIN
Abstract:	In the digital era, social media plays a vital role as a communication tool where users can share information and connect, with Instagram being one of the platforms. This study aims to determine whether Instagram content (X) directly or indirectly affects brand awareness (Y) through customer feedback (Z) as an intervening variable. The study population consists of Instagram followers of @minisoxsamono_kudus. Sampling was conducted using reliance available sampling through a questionnaire, resulting in 128 respondents. Data from the questionnaire were analyzed using regression and path analysis with SPSS 25 software. The results show that Instagram Content (X) significantly affects Customer Feedback (Z) with a t-value (12.604) > t-table (1.979) and Brand Awareness (Y) with a t-value (11.302) > t-table (1.979). Additionally, Customer Feedback (Z) significantly influences Brand Awareness (Y) with t-value (10.939) > t-table (1.979). Path analysis reveals that the direct effect of X → Y is 0.747, while the indirect effect of X → Z → Y is 1.2314. The Sobel test for mediation shows a t-value (8.323) > t-table (1.979), indicating that Instagram Content (X), through Customer Feedback (Z), has a positive and significant impact on Brand Awareness (Y). It is recommended that Miniso X Samono Kudus continue innovating and creatively improving the quality of Instagram content while consistently considering customer feedback to enhance the brand awareness of Miniso X Samono Kudus.
Keywords:	Content Instagram, Customer Feedback, Brand Awareness, Miniso X Samono
Source:	Journal of Theoretical and Applied Information Technology 15^th March 2025 -- Vol. 103. No. 5-- 2025
Full Text

Title:	ENHANCING RECURRENT NEURAL NETWORK PERFORMANCE FOR LATENT AUTOIMMUNE DIABETES DETECTION (LADA) USING EXOCOETIDAE OPTIMIZATION
Author:	B. SUCHITRA , R. KARTHIKEYAN , J. RAMKUMAR , V. VALARMATHI
Abstract:	Latent Autoimmune Diabetes in Adults (LADA) is frequently misclassified as type 2 diabetes, a significant issue highlighted in current diagnostic research. This misclassification, due to overlapping clinical features with other diabetes types, leads to suboptimal patient management. Existing models lack the precision to discern the subtle, time-dependent variations characteristic of LADA, underscoring a critical gap in diabetes diagnostics. This study introduces a groundbreaking approach, the Exocoetidae Optimization-Inspired Recurrent Neural Network (EO-RNN), which innovatively incorporates the adaptive flight patterns of Exocoetidae (flying fish) into optimizing RNNs. This bio-inspired algorithm enhances parameter tuning and model adaptability, significantly improving the detection accuracy of complex disease patterns. Employing the EO-RNN on the Gestational Diabetes Mellitus (GDM) dataset resulted in a classification accuracy of 95.36% and an F-measure of 95.47%, surpassing traditional models by over 30%. These results not only bridge the identified literature gap by providing a more effective diagnostic tool but also contribute substantial new knowledge about the integration of nature-inspired algorithms in medical diagnostics. The success of EO-RNN in enhancing diagnostic accuracy demonstrates its potential as a transformative tool in healthcare, promising for broader applications in detecting complex diseases where similar misclassification issues exist.
Keywords:	LADA Classification, Exocoetidae Optimization, Recurrent Neural Networks, Bio-Inspired Optimization, Diabetes Prediction, Machine Learning In Healthcare, Temporal Dependency Modeling
Source:	Journal of Theoretical and Applied Information Technology 15^th March 2025 -- Vol. 103. No. 5-- 2025
Full Text

Title:	DESIGN A CIRCUIT LEVEL LOW LEAKAGE AND ROBUST FLIP LECTOR APPROACH FOR SRAM CELL
Author:	DEEPAK MITTAL , V.K. TOMAR
Abstract:	In static random access memory (SRAM) cells, leakage power, stability, and speed have become significant challenges with the scale-down of technology. This paper presents a proposed Flip LECTOR approach. In the LECTOR approach, a leakage-controlled transistor (LCT) is always near its "cut-off" voltage for any input supply. In the GALEOR approach, gated transistors NMOS and PMOS are unsuitable for passing the VDD and ground at pullup (PUN) and pulldown (PDN), respectively. In MTCMOS, gated transistors may fail the gates, which reduces the noise margin. Also, due to additional mask layers, the fabrication process becomes complex. Hence, due to the above approaches, the operation of the 6T SRAM cell was vitiated. There is also signal quality contention, but this method does not have these problems because the adjacent LCT is always close to its linear voltage when one of the transistors in the SRAM cell is either PMOS or NMOS and "ON." LCT increases the resistance, which reduces leakage power. When compared to a 6T SRAM, LECTOR, GALEOR, and MTCMOS at 1V, the proposed method reduces the leakage power 2.55×, 5.84×, 1.08×, and 1.52× during read operation respectively. It is also 1.57×, 24×, and 1.15× improved during the write operation. The read delay is 3.82× and 3.83× smaller than that of LECTOR and GALEOR. The write delay is 1.21×, 1.26×, and 1.37× less than that of LECTOR, GALEOR, and MTCMOS. Proposed approach WSNM is 1.92×, 1.03×, 1.78×, and 1.43× better than 6T SRAM, LECTOR, GALEOR, and MTCMOS.
Keywords:	SRAM cell, leakage power dissipation, stability, power delay product, slew rate, Flip LECTOR Approach.
Source:	Journal of Theoretical and Applied Information Technology 15^th March 2025 -- Vol. 103. No. 5-- 2025
Full Text

Title:	EARLY-STAGE DETECTION AND CLASSIFICATION OF ALZHEIMER’S DISEASE USING CUTTING EDGE TECHNOLOGIES
Author:	NASSER NAMMAS ALBOGAMI
Abstract:	Alzheimer's disease has emerged as a significant global health issue in recent decades. However, low- and middle-income countries have been slow to acknowledge its severity. According to reports, the global prevalence of Alzheimer's disease is expected to exceed 55 million individuals by the end of 2023, with projections suggesting this figure will triple by 2050. Early detection and appropriate treatment plans can mitigate the risks associated with Alzheimer's disease. Yet, only 10 percent of affected individuals in developing nations receive a formal diagnosis from healthcare professionals. In this study, we propose a novel computer-aided diagnostic (CAD) based Alzheimer's disease detection and classification system that provides precise classification of the disease's current stage. The proposed scheme assists medical professionals to diagnose the Alzheimer's disease with maximum accuracy level 99.8%. Furthermore, the proposed approach improves the existing neuroimaging datasets by conducting analysis and preprocessing techniques such as annotation and labeling. To best of our knowledge, our proposed scheme is the the most appropriate and feasible approach that alleviate the timely detection of Alzheimer's disease, enabling patients to receive appropriate medical treatment and slow the progression of the condition.
Keywords:	Artificial Intelligence; Deep learning; Alzheimer's disease (AD); computer-aided diagnostic (CAD).
Source:	Journal of Theoretical and Applied Information Technology 15^th March 2025 -- Vol. 103. No. 5-- 2025
Full Text

Title:	IMAGE PROCESSING AND DEEP LEARNING FOR EXCELLENT ASPHALT CRACK DETECTION
Author:	BELAL A. ELMONEM , OSAMA E.EMAM , HELAL A.SULEIMAN
Abstract:	This study addresses the critical task of asphalt crack detection, essential for efficient road maintenance and infrastructure management. Traditional methods using raw road surface images often suffer from low detection accuracy under varied conditions. To overcome these limitations, our approach integrates advanced image-processing techniques that enhance input image quality prior to training, thus improving model generalization across diverse road conditions. This method significantly boosts detection performance, offering a reliable solution for civil engineering applications. The initial accuracy rates were 74.54% for Unet and 91.45% for CNN, which improved post-processing to 97.58% and 100%, respectively.
Keywords:	Crack Detection; Image Preprocessing; CrackSense; Asphalt Segmentation; Deep Learning; CNN (Convolutional Neural Network); UNet Architecture
Source:	Journal of Theoretical and Applied Information Technology 15^th March 2025 -- Vol. 103. No. 5-- 2025
Full Text

Title:	AN EXPLORATION OF 2D AND 3D CT SCAN IMAGES FOR DETECTING COVID-19 USING SPATIAL CONVOLUTIONAL NEURAL NETWORK
Author:	K. KIRUBANANTHAVALLI , Dr. P. SUNDARESWARAN
Abstract:	COVID-19 has been the most impactful pandemic in modern history, affecting people all across the world. The identification of COVID-19 relies mostly on lung Computer Tomography (CT) images. A computer-aided diagnosis (CAD) method for classifying COVID-19 CT images is presented in this study. In this study, assess the accuracy of 2D and 3D SCNN models that share similar architectural details. The dataset is sourced from the 2D and 3D CT scan images Dataset, Chest CT scans with COVID-19 related findings database. Threshold segmentation is the best method for separating the chest from the rest of the CT scan. An advanced collection of deep learning models, SCNN combines the best 2D and 3D systems. It combines slice-level assessments, a CNN model, and unique preprocessing and attention components. In light of this, the suggested study introduces SCNN, an image-processing-based COVID-19 detection model. The COVID-19 CT images used to train this model were split into three categories: COVID-19, pneumonia, and healthy subjects. If the input image does not include the required attributes, an image preprocessing pipeline may be used to extract the ROI. As part of the proposal, combine the predictions made by 2D and 3D SCNN models. Using contrastive learning and an attention mechanism, this work presents a classification method. By reducing the distance between images in the same category, contrastive learning may increase the feature space used for classification. To aid with classification, an attention mechanism may draw focus on a key area of the image while offering a visual representation of that area. We showed a significant increase in classification accuracy by studying SCNN classification. In addition, we have achieved a comprehensive visual representation as compared to conventional methods.
Keywords:	COVID-19, 2D and 3D spatial Convolutional Neural Network, Classification, Accuracy
Source:	Journal of Theoretical and Applied Information Technology 15^th March 2025 -- Vol. 103. No. 5-- 2025
Full Text

Title:	IMPROVING STUDENT ENGAGEMENT IN LEARNING REQUIREMENT ENGINEERING SUBJECT USING PAIR LEARNING
Author:	NURAZLINA MD SANUSI , ABDUL KARIM MOHAMAD
Abstract:	Requirement engineering is one of the disciplines in software engineering areas that play an important role in determining successful software development. Many researchers have highlighted the importance of requirement engineering aspects in software engineering. They pointed out that one of the difficulties of teaching requirement subjects is the preparation in the classroom to teach requirement engineering subject and make students engage. Learning requirement subject can be difficult for some students in the classroom. In this paper, we presented a new engagement framework using pair work learning in the classroom. We adapted pair work and explored this approach in teaching and learning requirement engineering subject. With the assistance of the learning management system (Moodle platform – in our university, we called ULearn), the activities and assessments designed in pair, we lead the students to linkage with engagement and lead them to learn.
Keywords:	Software Engineering, Student Engagement, Pair Learning, Requirement Engineering, Requirement Engineering Education.
Source:	Journal of Theoretical and Applied Information Technology 15^th March 2025 -- Vol. 103. No. 5-- 2025
Full Text

Title:	DEVELOPING AN IT WORKFORCE PLANNING WITH CROWDWORKING MODEL FOR INDONESIAN BANKING
Author:	CHRISTIAN HADI , HARJANTO PRABOWO , HARCO LESLIE HENDRIC SPITS WARNAR , JUNEMAN ABRAHAM
Abstract:	This study proposes an IT workforce planning with crowdworking model for Indonesian banking industry, addressing challenges on IT workforce planning process and rising trend of crowdworking. Using a Design Science Research Method, we developed a model based on literature review, interviews and FGD with expert practitioners in Indonesian banking industry and has been evaluated through user acceptance test and expert judgment. The proposed model incorporates six key features: analyse current workforce, future workforce estimation, gap analysis, action identification, and process evaluation, in addition we are also adding internal crowdworking components as one of strategy to improve IT workforce productivity. A use case diagram, class diagram, and prototype were developed to demonstrate the model's practical application. Findings suggest that the model can improve IT workforce planning process, particularly in aligning HR with business objectives and facilitating crowdworking process. This study contributes to both theory and practice by providing a structured approach to IT workforce planning with crowdworking in Indonesian banking industry.
Keywords:	Crowdworking, Internal Crowdworking, IT Workforce Planning, Workforce Planning
Source:	Journal of Theoretical and Applied Information Technology 15^th March 2025 -- Vol. 103. No. 5-- 2025
Full Text

Title:	OPTIMIZED FREQUENT SUBGRAPH MINING USING ITERATIVE MAPREDUCE FOR ENHANCED SCALABILITY AND PERFORMANCE
Author:	NAGA MALLIK ATCHA, JAGANNADHA RAO D B, VIJAYAKUMAR POLEPALLY
Abstract:	Frequent subgraph mining (FSM) is a core graph analysis task arising from many application domains, including bioinformatics, chemoinformatics, and social network analysis. Traditional FSM methods are not scalable to large datasets due to in-memory computations and inefficient candidate pruning, as found in gSpan and Apriori-based techniques. Although some recently distributed approaches, such as G-thinker and FlexMiner, have tried to overcome them, they are still confronted with high computational overhead, excessive data shuffling, and scalability. Such problems necessitate a sound, scalable approach for large-scale graph mining in the modern era. This research proposes a novel, sophisticated framework, and algorithm called Frequent Subgraph Mining Using MapReduce (FSM-MR) with inherent optimizations. This efficient algorithm incorporates mapper combiners for minimizing data shuffling, canonical labeling for avoiding repeated enumeration of identical subgraphs, and dynamic support thresholds for effective pruning. FSM-MR, implemented in a Hadoop environment, shows better performance with up to 50% runtimes shorter than the state-of-the-art methods, with near-linear scalability with the addition of cluster nodes. The ability of the proposed framework to process large-scale graph datasets makes it beneficial for applications involving scalable, efficient graph mining. FSM-MR overcomes methodological limitations in the current state-of-the-art algorithms, helping set the stage for future research in these areas and fostering graph analytics in various scientific and industrial settings.
Keywords:	Frequent Subgraph Mining, MapReduce Framework, Scalability, Distributed Graph Analysis, Big Data Processing
Source:	Journal of Theoretical and Applied Information Technology 15^th March 2025 -- Vol. 103. No. 5-- 2025
Full Text

Title:	LEVERAGING COUNT VECTORIZER FOR JOB TITLE PREDICTION: A COMPARATIVE STUDY OF MACHINE LEARNING ALGORITHMS
Author:	D.DEEPTHI , KAMINENI B.T.SUNDARI , B.SATISH BABU , K.THRILOCHANA , A.MAHALAKSHMI , SNEHA.H.DHORIA
Abstract:	Job title prediction from description data is a crucial task in automating job classification and improving digital job search platforms. This study evaluates the performance of advanced machine learning models—K-Nearest Neighbors, Support Vector Machines, Gradient Boosting, and Logistic Regression—for predicting job titles based on descriptive text. Experiments were conducted using two data splits: 70-30 and 80-20 for training and testing. Results reveal that the 70-30 split consistently outperforms the 80-20 configuration in terms of prediction accuracy. Among the evaluated models, Gradient Boosting achieved the highest performance, with an accuracy of 98.05%, utilizing the Count Vectorizer method. Furthermore, Gradient Boosting recorded the highest F1-score of 0.88, along with a recall of 0.81 for the class for 70-30. These findings highlight the superior capability of Gradient Boosting in capturing complex patterns in textual data and emphasize the significance of data pre-processing and splitting strategies. The outcomes contribute to the optimization of machine learning applications in employment platforms, enhancing user experience and efficiency in matching candidates with appropriate job opportunities. This paper focuses on the gap in bringing job seekers and proper opportunities to improve a more transparent, efficient, and trustworthy job marketplace.
Keywords:	Job title, Machine learning, Job descriptions, Gradient Boosting, Count Vectorizer, F1-score.
Source:	Journal of Theoretical and Applied Information Technology 15^th March 2025 -- Vol. 103. No. 5-- 2025
Full Text

Title:	A NOVEL FUZZY RULE BASED DECISION-MAKING APPROACH FOR TASK OFFLOADING ON MULTI-TIER MOBILE-EDGE CLOUD COMPUTING SYSTEMS
Author:	R. AISHWARYA, G. MATHIVANAN
Abstract:	Mobile Edge Cloud Computing (MECC) is a developing distributed processing method, which delivers access to CC services at the network’s edge and closer mobile operators. In a speedily varying and dynamic environment, it is very cruel to discover the optimum target server for managing unloaded tasks since we don’t recognize the end user’s demand further. By offloading tasks at the network's edge rather than transmitting them to a remote cloud, MECC can recognize flexibility and actual handling. In the present study, the varying desires of Internet of Things (IoT) applications at dissimilar phases are often neglected in the context of computation offloading. This study introduces fuzzy rule-based decision-making for task offloading approach (FRBDM-TOLA) technique on multi-tier MECC systems. Initially, the presented FRBDM-TOLA technique uses fuzzy clustering algorithms like Fuzzy C-Means (FCM) to categorize tasks into fuzzy subcategories based on their attributes, intensity levels, and temporal aspects. Moreover, the developed method employs the Hybrid Fuzzy-Neural Network (HFNN) approach to select the most suitable target node for the offloading of tasks depending on latency sensitivity function, server capacity, and the state of the network. The HFNN is a hybrid of the FL and the NN used for the rule generation with reference to the intensity level and traffic flow. FL can claim linguistic comments and uncertainties and neural networks can take up complex patterns from the data offered to them. With the aim to improve the performance of the HFNN method the FRBDM-TOLA technique employs Spotted Hyena Optimizer (SHO) for hyperparameter optimization. The developed models optimally employ processes of classification and clustering so as to increase classification accuracy of evaluating network traffic data.
Keywords:	Fuzzy Logic, Internet of Things, Mobile-Edge Cloud Computing, Spotted Hyena Optimizer, Task Offloading.
Source:	Journal of Theoretical and Applied Information Technology 15^th March 2025 -- Vol. 103. No. 5-- 2025
Full Text

Title:	A NOVEL INTEGRATION OF MACHINE LEARNING -BASED DATA CLASSIFICATION WITH OPTIMIZED CRYPTOGRAPHIC TECHNIQUES FOR SECURE CLOUD STORAGE
Author:	BHARATH KUMAR RAMA ,DR. S. THAIYALNAYAKI
Abstract:	Security measures for cloud storage remain crucial because modern data management depends heavily on such storage, yet it exposes vulnerable data points. The proposed hybrid security framework combines machine learning methodologies with cryptographic protocols to improve cloud environment data protection. The proposed system utilizes Support Vector Machine (SVM) and Random Forest (RF) machine learning classifiers that sort data into three security levels: public, private, and confidential through established security attribute definitions. We employ specific encryption techniques, which include RSA and AES for moderately sensitive data, whereas ChaCha20-Poly1305 encrypts high-security sensitive information. Data integrity protection through the ChaCha20-Poly1305 scheme exists because it combines encryption with authentication that adds a message authentication code (MAC) derived from the ciphertext and keystream. The selective encryption approach achieves better computational speed because it avoids unnecessary waste of processing resources through complex encryption protocols. Testing cloud-stored databases showed our combination method delivers faster performance during encryption time of 5-10% improvement than independent ChaCha20 implementation. This framework delivers improved scalability as well as security-efficiency balances that make it appropriate for actual cloud storage applications.
Keywords:	Data security, cloud computing, data classification, machine learning, ChaCha20
Source:	Journal of Theoretical and Applied Information Technology 15^th March 2025 -- Vol. 103. No. 5-- 2025
Full Text

Title:	AN EXPLORATION OF 2D AND 3D CT SCAN IMAGES FOR DETECTING COVID-19 USING SPATIAL CONVOLUTIONAL NEURAL NETWORK
Author:	K. KIRUBANANTHAVALLI , Dr. P. SUNDARESWARAN
Abstract:	COVID-19 has been the most impactful pandemic in modern history, affecting people all across the world. The identification of COVID-19 relies mostly on lung Computer Tomography (CT) images. A computer-aided diagnosis (CAD) method for classifying COVID-19 CT images is presented in this study. In this study, assess the accuracy of 2D and 3D SCNN models that share similar architectural details. The dataset is sourced from the 2D and 3D CT scan images Dataset, Chest CT scans with COVID-19 related findings database. Threshold segmentation is the best method for separating the chest from the rest of the CT scan. An advanced collection of deep learning models, SCNN combines the best 2D and 3D systems. It combines slice-level assessments, a CNN model, and unique preprocessing and attention components. In light of this, the suggested study introduces SCNN, an image-processing-based COVID-19 detection model. The COVID-19 CT images used to train this model were split into three categories: COVID-19, pneumonia, and healthy subjects. If the input image does not include the required attributes, an image preprocessing pipeline may be used to extract the ROI. As part of the proposal, combine the predictions made by 2D and 3D SCNN models. Using contrastive learning and an attention mechanism, this work presents a classification method. By reducing the distance between images in the same category, contrastive learning may increase the feature space used for classification. To aid with classification, an attention mechanism may draw focus on a key area of the image while offering a visual representation of that area. We showed a significant increase in classification accuracy by studying SCNN classification. In addition, we have achieved a comprehensive visual representation as compared to conventional methods.
Keywords:	COVID-19, 2D and 3D spatial Convolutional Neural Network, Classification, Accuracy
Source:	Journal of Theoretical and Applied Information Technology 15^th March 2025 -- Vol. 103. No. 5-- 2025
Full Text

Title:	CHATBOT ADOPTION IN BANKING: ENHANCING CUSTOMER EXPERIENCE AND SATISFACTION IN INDONESIA – JAKARTA URBAN AREA
Author:	NYI MAS ZAHRA ALMIRA , AFRA FAIRUZ RAMADHANI , ESTY NOVITA RAHMAN , JERRY S. JUSTIANTO
Abstract:	The widespread adoption of chatbots in the banking industry has transformed customer service delivery. The widespread adoption of chatbots in the banking industry has transformed customer service delivery. However, there is limited research on the impact of chatbots on customer experience and satisfaction, particularly in emerging markets. This study aims to fill this gap by investigating how the adoption of chatbots in the Indonesian banking sector influences online customer experience and satisfaction, with a focus on usability and security features. The research employed a quantitative design, using a structured questionnaire to collect data from 360 banking customers in the Jakarta area who have experience using chatbot services. Structural Equation Modeling (SEM) was used to analyze the relationships between chatbot adoption, online customer experience, and customer satisfaction. The findings revealed that the adoption of chatbots significantly influences online customer experience, which plays a crucial mediating role between chatbot adoption and customer satisfaction. This highlights the importance of enhancing online customer experience to improve customer satisfaction through chatbot implementation. Correspondingly, the study contributes to the existing literature by addressing a notable gap in understanding the security aspects of chatbot adoption. It provides a comprehensive framework that integrates security considerations with usability features. This offers valuable insights for banks implementing chatbot technologies while ensuring customer trust and data protection.
Keywords:	Chatbot Adoption, Banking Industry, Customer Experience, Customer Satisfaction, Security
Source:	Journal of Theoretical and Applied Information Technology 15^th March 2025 -- Vol. 103. No. 5-- 2025
Full Text

Title:	LEAFDISEASENET: A NOVEL CONVOLUTIONAL NEURAL NETWORK ARCHITECTURE FOR DETECTING LEAF DISEASES IN AGRICULTURAL CROPS
Author:	CH.SANTOSHI, SURESH KUMAR MANDALA
Abstract:	For productive agriculture and healthy crops, plant disease detection is essential. Conventional disease detection techniques are human, arbitrary, and error-prone and require automated solutions. Specifically, convolutional neural networks (CNNs) are deep learning models that have demonstrated great promise for identifying plant diseases; however, they are primarily model-specific for a particular kind of plant and disease, restricting them from generalizability over diverse crops and conditions. Moreover, class imbalance, environmental noise, and variability of disease symptoms make it even harder for these models to perform. In response to these problems, we present a sophisticated, integrated deep-learning framework called LeafDiseaseNet. Our proposed architecture includes a squeeze-and-excitation (SE) block with residual connections and data augmentation, enabling the model to be generalized for different plant diseases and environmental conditions. LeafDiseaseNet is evaluated on the Plant Village dataset and yields 97.68%, which outperforms current best practices regarding precision and resilience. It also exhibits good quality regarding high imbalanced class recognition and absolute environmental noise robustness, which may become promising approaches for implementing plant disease recognition in real-world agriculture situations. The framework aims to provide an efficient and scalable strategy for automating plant disease detection that can play a massive part in precision farming by making information about the disease available promptly and with sufficient time for action.
Keywords:	Squeeze-And-Excitation, Deep Learning, Plant Disease Detection, Convolutional Neural Networks, and Residual Connections
Source:	Journal of Theoretical and Applied Information Technology 15^th March 2025 -- Vol. 103. No. 5-- 2025
Full Text

Title:	HOW DOES RISKS AND ETHICAL ISSUES IMPACT THE ADOPTION OF ARTIFICIAL INTELLIGENCE IN AUDITING?
Author:	ANDI MUHAMMAD ALIF FAHD ZAKY , RINDANG WIDURI
Abstract:	AI has become commonly used due to its many benefits. It is also applied in auditing, where AI has been utilized to support the auditor's auditing process. AI reputation as of now has been a mixed one where its seen as beneficial but also accompanied by concerns and debates. Applying AI in audits may result in risk or ethical issues. This study uses the extended TAM (Technology Acceptance Model) by adding Risk and Ethical Issues variables, hypothesized to affect Perceived Usefulness and Ease of Use. Data is collected by distributing questionnaires to auditors working in Indonesia's Big 4 auditing firm. Data is analysed through SEM-PLS. The result indicates that Risk and Ethical Issues does not affect Perceived Usefulness, but affect Perceived Ease of Use negatively. The perceived usefulness and ease of use influence the intention to use, while the perceived ease of use influences perceived usefulness. Risk and ethical issues exist, albeit they do not significantly impact how auditor perceived their benefits. It does, however, makes it harder to use. As such, auditors should thread carefully on utilizing AI in future endeavours. Previous research have tackle the topic of the use and ethical concerns of AI in auditing. This research discusses things that have not been addressed in previous research, namely how the risks and ethical issues that arise with AI influence auditors' perceptions of AI in audits and their willingness to use them.
Keywords:	Risk, Ethical Issues, Artificial Intelligence (AI), TAM (Technology Acceptance Model), Innovation.
Source:	Journal of Theoretical and Applied Information Technology 15^th March 2025 -- Vol. 103. No. 5-- 2025
Full Text

Title:	LEARNING-BASED APPROACH FOR AUTOMATIC SEGMENTATION OF IMAGES
Author:	M SRIVIDYA . DR. VENUBABU RACHAPUDI
Abstract:	One of the key fields of study for computer vision applications is image segmentation. Often, segmentation is used as a preprocessing step in various real-world applications. The existing literature has revealed that image segmentation can be done with heuristic and learning-based techniques. With the emergence of artificial intelligence, learning-based approaches became popular in computer vision applications. However, there is a need to explore ML and DL-based methods with optimizations geared towards image segmentation efficiency. In this document, we suggested an algorithm known as Spectral Clustering-Based Image Segmentation (SCB-IS), which exploits the spectral clustering process for efficient image segmentation. We also proposed an approach to segmentation based on deep learning that exploits SegNet, which is widely used for image segmentation. We proposed another algorithm known as Deep Learning-Based Image Segmentation (DLB-IS). Our empirical study with benchmark datasets revealed that the proposed deep learning-based algorithm outperforms many cutting-edge deep learning models, featuring the highest accuracy of 88.50%. The suggested learning-based approaches can be integrated into real-world computer vision applications for efficient segmentation of images.
Keywords:	Image Segmentation, Machine Learning, Deep Learning, Clustering, Artificial Intelligence, SegNet
Source:	Journal of Theoretical and Applied Information Technology 15^th March 2025 -- Vol. 103. No. 5-- 2025
Full Text

Title:	THE INTELLIGENT COLLABORATIVE SUPPLY CHAIN MANAGEMENT WITH LARGE LANGUAGE MODELS
Author:	SIRILUK PHUENGROD , PANITA WANNAPIROON , PRACHYANUN NILSOOK
Abstract:	This research explores the development of an Intelligent Collaborative Supply Chain Management (iCSCM) system, driven by Large Language Models (LLMs), to enhance operational efficiency and facilitate collaboration among academic, governmental, and private sectors in advancing University Holding Companies (UHCs). Despite significant progress in AI-driven supply chain management, challenges remain in effectively aligning academic research with industry demands, resulting in suboptimal resource utilization and missed opportunities for commercialization. This study seeks to address these issues by proposing an AI-driven framework to optimize collaboration within this ecosystem. Employing system design frameworks, architectural evaluation matrices, and expert surveys, the study evaluates the proposed system’s effectiveness, demonstrating a high level of suitability (Mean = 4.73, SD = 0.30). The findings underscore the transformative potential of large language models in enhancing collaborative supply chain processes, equipping universities to serve as key innovation hubs that bridge the gap between research and industry applications.
Keywords:	Intelligent Collaborative Supply Chain Management, Large Language Models, Artificial Intelligence, University Holding Company
Source:	Journal of Theoretical and Applied Information Technology 15^th March 2025 -- Vol. 103. No. 5-- 2025
Full Text

Title:	EXPLORING THE KEY DRIVERS BEHIND THE ADOPTION OF COMPUTER-ASSISTED AUDIT TECHNIQUES IN HIGHER EDUCATION
Author:	FITTRYZKA ZIE SHAKIRA , ANANDA LUSI BRILIANA , ROCHANIA AYU YUNANDA
Abstract:	The utilization of computers in organizations for information processing grows, and it becomes increasingly essential for auditors to adopt computerized auditing techniques. This study investigates students' views on the factors that affect their behavioral intentions regarding Computer-Assisted Audit Techniques (CAAT) in the academic sector. The quantitative research relies on primary data collected via questionnaires distributed to the research subjects, specifically active students in Accounting and Finance programs currently enrolled in Auditing courses at various universities in Indonesia. A total of 228 participants from 35 universities took part in this research. The selection of participants was conducted using a simple random sampling method. This research uses the PLS Algorithm and Bootstrapping method to examine the relationships between various variables. The independent variables in this research include Effort Expectations, Social Influence, Performance Expectations, Facilitating Conditions, and Trust. While, Satisfaction serves as the mediating variable, and Behavioral Intention is identified as the dependent variable. The proposed model in this research consists of nine hypotheses, of which six have been validated while three remain unvalidated. Previous studies indicate that only some universities incorporate audit software into their courses due to inadequate infrastructure, highlighting a gap in technological competencies in accounting and auditing education. This research contributes to the broader discourse on advancing these competencies, ultimately preparing students to meet the dynamic demands of the global audit environment and enhancing their readiness for professional work as practising auditors. The findings provide insights into student expectations regarding the Computer-Assisted Audit Techniques (CAATs) course, thereby enhancing the overall quality of effective learning within the course.
Keywords:	Computer Assisted Audit Techniques, Technology, Education, Learning, UTAUT model
Source:	Journal of Theoretical and Applied Information Technology 15^th March 2025 -- Vol. 103. No. 5-- 2025
Full Text

Title:	A DEGREE CONSTRAINED RESTRICTED MINIMUM SPANNING TREE METHOD FOR ROBOT NAVIGATION WITH ENHANCEMENT IN COLLISION AVOIDANCE
Author:	D SATYANARAYANA , ABDULLAH SAID Al KALBANI , GOPAL RATHINAM , NADIR KAMAL SALIH IDRIES , ALAYA HAMYAR AL AZZANI
Abstract:	Robotics is the process of designing and building robots. The robot has to be intelligent enough to manage its tasks on its own. The mobile robot needs to be smart for travelling from a source to a destination location. The path identification of a mobile robot is a challenging task for the robots. In this paper, we propose a new localized algorithm, which uses Radio Frequency Identification technology to create a path of the mobile robot from a source to a specific destination. The proposed method constructs a route map, which is based on limiting the node degree of a minimum spanning tree. This method is also useful to reduce robot collision avoidance during the mobile robot travelling process. The simulation results show the performance of the proposed method.
Keywords:	Robotics, Mobile Robot, Minimum Spanning Tree, RFID Tags, Path Identification, Collision Avoidance.
Source:	Journal of Theoretical and Applied Information Technology 15^th March 2025 -- Vol. 103. No. 5-- 2025
Full Text

Title:	BEHAVIORAL PROFILING FOR CARD-NOT-PRESENT FRAUD DETECTION LEVERAGING ISO8583 DATA TO IDENTIFY ANOMALOUS PATTERNS
Author:	MAROUANE AIT SAID , ABDELMAJID HAJAMI , AYOUB KRARI
Abstract:	Card-Not-Present (CNP) fraud continues to rise, with fraudsters exploiting sensitive cardholder data to execute unauthorized transactions. This paper presents a behavioral profiling framework that uses ISO8583 fields to identify transaction anomalies indicative of fraudulent activity. By analyzing fields such as transaction amounts, merchant categories, POS entry modes, and terminal identifiers, the framework establishes behavioral baselines for individual cardholders and aggregates patterns across similar cardholder pro-files. Fraudulent behaviors, such as testing cards with small transactions before escalating to larger amounts, are detected by monitoring deviations from typical spending patterns. These deviations are flagged as anomalies, enabling early detection and prevention of fraudulent activities. The proposed framework also considers shared behavioral insights across multiple cardholders to enhance detection accuracy while minimizing false positives. A prototype implementation demonstrates the practical applicability of this approach, offering a scalable and efficient solution for CNP fraud detection using ISO8583 data. By focusing on behavioral profiling, this work bridges the gap between traditional rule-based systems and adaptive, data-driven fraud prevention methods.
Keywords:	Card-Not-Present Fraud, Behavioral Profiling, Fraud Detection Framework, Payment Systems Security, Merchant Category Code (MCC).
Source:	Journal of Theoretical and Applied Information Technology 15^th March 2025 -- Vol. 103. No. 5-- 2025
Full Text

Title:	A MODIFIED K-MEANS APPROACH FOR EFFECTIVE CLUSTERING USING WEIGHTED ADJACENT MATRIX
Author:	HEMA BHARDWAJ , DR. D. SRINIVASA RAO
Abstract:	K-means clustering has several limitations, such as sensitivity to initialization and determining the number of clusters. It is sensitive to outliers, especially when identifying clusters with irregular shapes or varying sizes. Handling categorical data directly in k-means can be challenging. This study aims to present methods to improve the existing k-means clustering algorithms. It proposes designing two distinct proximity matrices for this purpose. The study suggests that the new algorithm performs better than traditional clustering methods based on several evaluation metrics. Randomly chosen centroids lead to unstable outcomes. The unpredictable initialization of centroids makes it difficult to replicate clustering results. Spectral clustering begins by creating a similarity matrix, followed by eigenvalue decomposition applied to the Laplacian matrix. This decomposition results in a spectral representation. However, optimal clustering outcomes cannot be guaranteed in the initial stage of the spectral clustering algorithm. This research proposes a solution to this issue. An Initialization & Similarity approach is recommended, where both the representation and the similarity matrix are determined in a cohesive manner. Additionally, it improves clustering performance by using sum of norms regularization. Based on evaluation metrics, this clustering technique proves to be better than the original k-means algorithm. Using normalized mutual information, purity, and accuracy as measures, the proposed technique demonstrates superiority over traditional algorithms. This study presents a novel approach to K-Means clustering by integrating a weighted adjacent matrix, significantly enhancing clustering accuracy and effectively handling high-dimensional data. The proposed methods, KM-AM and KM-WAM, show improved performance metrics such as normalized mutual information, accuracy, and purity, offering a more efficient and robust solution for various data analysis applications.
Keywords:	K-means clustering, similarity matrix, spectral clustering, laplacian matrix
Source:	Journal of Theoretical and Applied Information Technology 15^th March 2025 -- Vol. 103. No. 5-- 2025
Full Text

Title:	A SYSTEMATIC LITERATURE REVIEW ON BIBLIOMETRIC PROFILING ANALYSIS USING MACHINE LEARNING
Author:	ANISA IRSYAHIDAH A. RIDZUAN , W. M. AMIR FAZAMIN W. HAMZAH , MOKHAIRI MAKHTAR , MUMTAZIMAH MOHAMAD , ISMAHAFEZI ISMAIL , AZLIZA YACOB
Abstract:	In the constantly expanding area of scholarly inquiry, the need for efficient instruments or tools to measure and evaluate the effectiveness of literary works cannot be overestimated. As a field of study, bibliometrics is truly informative as it measures research output in a given field, further assisting in the areas of research grants, publications and evidence for decision-making and future planning in the education sector. It is a systematic way of evaluating productivity. Hence, this research attempts to carry out a literature review regarding the application of Bibliometric Profiling Analysis in the context of Machine Learning (ML). This research involved the assessment and evaluation of the research grants, innovations, published articles, journals and proceeding papers. All materials were retrieved using online databases such as IEEE Xplore, Ingenta Connect, Sage Journal, Science Direct and Proquest. To achieve this, technologies including data mining, network analysis and natural language processing techniques were applied. From the review, the Support Vector Machine (SVM) is a machine learning algorithm frequently used in Bibliometric Profiling Analysis.
Keywords:	Bibliometrics Profiling Analysis, Machine Learning, Natural Language Processing
Source:	Journal of Theoretical and Applied Information Technology 15^th March 2025 -- Vol. 103. No. 5-- 2025
Full Text

Title:	AN INTUITIVE ANALYSIS ON EARLY DETECTION WITH IAGT MODEL FOR COTTON CROP YIELD PREDICTION
Author:	PORANDLA SRINIVAS , DR. SURESH A
Abstract:	Cotton yield prediction plays a critical role in modern agriculture, influencing food security, economic stability, and effective resource management. Despite the advancements in various predictive algorithms like support vector machines (SVM), random forests, and artificial neural networks, challenges persist in handling the complex, high-dimensional, and non-linear nature of agricultural data. Traditional models struggle with issues such as dynamic environmental fluctuations, incomplete datasets, and an inability to effectively adapt to evolving conditions in the field. As a result, these models often fail to provide accurate and timely predictions, leading to inefficiencies in crop management, resource allocation, and risk mitigation. This study addresses the knowledge gap in cotton yield prediction by introducing the Integrated Adaptive Growth Tree (IAGT) algorithm, which combines decision trees with deep learning techniques for real-time adaptation to changing agricultural conditions. By integrating multi-source data, including satellite imagery, weather forecasts, and soil sensor readings, the IAGT model offers a novel approach to yield forecasting, surpassing traditional methods in both accuracy and adaptability. Simulations using both original and synthetic cotton yield datasets showed a remarkable 98% accuracy, demonstrating significant improvements in prediction performance. This study not only provides new insights into the effective integration of diverse data sources for crop yield forecasting but also introduces a robust framework for early-stage disease detection and anomaly identification, thus contributing to the growing field of precision agriculture. The IAGT model’s success in enhancing cotton yield predictions sets the stage for broader applications in crop management and agricultural sustainability.
Keywords:	Integrated Adaptive Growth Tree (IAGT), Convolution Neural Networks (CNN), Gated Adversarial Neural Networks (GAN), Support Vector machines (SVM),
Source:	Journal of Theoretical and Applied Information Technology 15^th March 2025 -- Vol. 103. No. 5-- 2025
Full Text

Title:	ANALYSIS OF ACUTE LYMPHOBLASTIC LEUKEMIA CLASSIFICATION WITH MACHINE AND DEEP LEARNING TECHNIQUES
Author:	B. REVATHI1, M. KALIAPPAN , E. MARIAPPAN , S.K. KEZIAL ELIZABETH
Abstract:	Leukemia is a term for a cancerous condition that affects the organs that produce blood. The inordinate batch of crude white blood cells (WBS) in the bone marrow is the cause of leukemia, a fatal disease. There is a greater possibility of recovery if leukemia is discovered early. In the treatment of leukemia, Deep Learning and Machine Learning algorithms are commonly employed, especially in detecting whether leukemia is progressing in patients. Conventional machine learning and deep learning approaches are practical guides in computer vision that improve the speed and accuracy of identifying and categorizing images (medical), including minuscule blood cells. This survey offers a thorough examination of the recognition and assortment of WBCs and acute leukemia in microscopic blood cells. In this survey, various directions of classification of Acute Lymphoblastic Leukemia (ALL), including the following steps in the methods: Data enhancement, preprocessing, segmentation, extracting features, selecting features(reduction), and classification, with an emphasis on the classification stage. The models using these classifiers have higher performance metrics when compared to the other model. Our suggestion is to offer for identifying and categorizing acute leukemia.
Keywords:	Leukemia, Classification, Medical Imaging, Cancer, Performance Metrics
Source:	Journal of Theoretical and Applied Information Technology 15^th March 2025 -- Vol. 103. No. 5-- 2025
Full Text

Title:	OPTIMIZED BRAIN MRI SEGMENTATION: K-MEANS++ AND VECTORISED FUZZY MEMBERSHIP COMPUTATION
Author:	R. LAKSHMI PRAVALLIKA , R. PRADEEP KUMAR REDDY
Abstract:	Magnetic Resonance Imaging (MRI) is a well applied method of brain analysis, because of its capability of acquiring detailed anatomical information. Accurate brain MRI segmentation is required for diagnosing brain related disorders. A brain MRI segmentation framework based on K-Means++ clustering and a novel vectorized fuzzy membership computation is proposed as the work that introduces an optimized solution to this problem with additional accuracy and speed. K-Means++ is deployed to initialize cluster centers in order to improve convergence time and quality of the segmentation. The computed vectorized fuzzy membership functions yield additional tissue segmentation for fine classification of tissues (gray matter, white matter and cerebrospinal fluid etc.) that are not provided by a tissue region segmented alone. A combined method based on robust noise reduction through application of Gaussian filtering linked to robust intensity normalization on an image quality basis to remove artifacts based upon noisy images is presented. Small, spurious regions are then removed using post-processing techniques. On benchmark brain MRI datasets, experiments demonstrate the superiority of the optimized segmentation method in terms of both segmentation accuracy and computational efficiency as compared to traditional K Means and Fuzzy C Means (FCM) algorithms, all with robustness against noise. The proposed method achieves 0.21% and 0.52% improvements in accuracy over traditional K-Means and FCM. Proposed method showing significant improvement in Dice Similarity Coefficient (DSC), Jaccard’s Index (JI), precision, recall, F1-score and MSE parameters compared to K-Means and FCM. This contribution makes a computationally efficient and more accurate hybrid segmentation approach to integrate K-means++ and vectorized fuzzy membership computation so as to boost the reliability of brain imaging analyses and clinical decision making.
Keywords:	Brain Tumor, Fuzzy C Means, Fuzzy membership functions, K-Means++, MRI Image.
Source:	Journal of Theoretical and Applied Information Technology 15^th March 2025 -- Vol. 103. No. 5-- 2025
Full Text

Title:	HOW CAN MODEL-DRIVEN ARCHITECTURE AUTOMATE DATA CATALOGS FOR ENHANCED DATA MANAGEMENT?
Author:	ASMAE BOUFASSIL , FADWA BOUHAFER , AMINE EL HADDADI , MOHAMED CHERRADI , ANASS EL HADDADI
Abstract:	In recent years, the scalability and maintenance of data catalogs posed significant challenges that hindered organizational efficiency and data accessibility. This paper examined these issues, highlighting the need for an automated approach. It advocated for the use of Model-Driven Architecture (MDA) to streamline the creation and maintenance of data catalogs. Through this approach, key data catalog components were automatically generated from higher-level models, minimizing manual work and improving both data integrity and system functionality. The findings indicated a considerable reduction in errors and operational demands, along with notable improvements in manageability and scalability. This integration of MDA into data catalog frameworks thus presented a compelling solution to the persistent challenges of data management, setting a new standard for efficiency and effectiveness in managing organizational data.
Keywords:	Data Catalog, Model-Driven Architecture (MDA), Data Management, Automated Data Catalog, Metadata Management.
Source:	Journal of Theoretical and Applied Information Technology 15^th March 2025 -- Vol. 103. No. 5-- 2025
Full Text

Title:	NOVEL PERFORMANCE ANALYSIS OF YOLOV5 AND YOLOV8 FOR CURCUMA LONGA LEAF DISEASE IDENTIFICATION
Author:	RANJITHKUMAR BANALA, RAJESH DUVVURU
Abstract:	The medical plant leaf disease detection plays a crucial role in the production of high-quality medicines globally. Particularly, the medicinal leaves used in Ayurveda medicine have significantly reduced the impact of COVID-19 in India and other parts of the world. Among such medicinal plants, the Curcuma longa, popularly known as turmeric, acts as an antibiotic that reduces the impact of lung-infected diseases. The current study concentrates on identifying early diseases in Curcuma longa leaves to enhance production, a significant challenge for farmers and practitioners. The current study pinpoints the most effective deep learning algorithms for distinguishing between the three classes of Curcuma longa: healthy leaves, leaf blotch, and leaf spot. The study employed the most popular and successful deep learning methods, such as Yolo V5 and V8, to identify the diseases for the first time on the 'Duggirala variant Curcuma longa Dataset', and achieved very high classification results. The mean average precision (mAP) results of YoloV5 have achieved 98.6%, whereas the YoloV8 method attained only 86.1%. We base the performance metrics of the two algorithms on training loss and validation loss, characterizing both training and validation losses with parameters like objectness (85%), box validation (96%), and classification analysis (98.2%). Based on our experimentation, we found that YoloV8 exhibits very high training loss and validation loss, while YoloV5 shows minimal losses. The experimentation results state that YoloV5 is best suited to detect the diseased and healthy classes for the novel Duggirala variant Curcuma longa Dataset. Overall, Yolov5 outperforms YoloV8 by 12.8 percent, and we recommend using the YoloV5 model for smart farming in turmeric plantations over other deep-learning models.
Keywords:	Deep Learning, Image Processing, Yolov5, Yolov8, Curcuma Longa Leaf Diseases.
Source:	Journal of Theoretical and Applied Information Technology 15^th March 2025 -- Vol. 103. No. 5-- 2025
Full Text

Title:	BOOSTING STUDENT PERFORMANCE PREDICTION IN E-LEARNING: A HYBRID FEATURE SELECTION AND MULTI-TIER ENSEMBLE MODELLING FRAMEWORK WITH FEDERATED LEARNING
Author:	N S KOTI MANI KUMAR TIRUMANADHAM , THAIYALNAYAKI S , NIRUPA V , M MADHAVI , PERURI VENKATA ANUSHA , V S PAVAN KUMAR , VAHIDUDDIN SHARIFF
Abstract:	This research introduces a novel and advanced methodology for predictive modeling using federated learning, addressing critical challenges such as data privacy, class imbalance, and model performance. Unlike traditional centralized approaches, our work ensures data privacy through federated learning, enabling high-performance models without exposing sensitive data. The novelty of our approach lies in the integration of advanced preprocessing techniques, such as the Synthetic Minority Oversampling Technique (SMOTE) for class imbalance, hybrid feature selection by the combination of Boruta algorithm and L2 regularization's (Boruta-L2) for robust feature selection, and a 3-tier ensemble model with cutting-edge hyperparameter tuning techniques, including Bayesian Optimization, Random Search, and Particle Swarm Optimization (PSO). As a result, our global model achieves an accuracy of 98.90%, significantly outperforming previous methodologies. The advancements in our work are highlighted by the superior model performance, scalability, and privacy preservation, making it a significant contribution to federated learning. This research provides a comprehensive, efficient, and privacy-preserving solution for distributed predictive tasks, setting a new benchmark in machine learning applications across various domains.
Keywords:	Boruta, L2 Regularization, Particle Swarm Optimization (PSO), E –Learning, Federated Learning, Hyperparameter Tuning.
Source:	Journal of Theoretical and Applied Information Technology 15^th March 2025 -- Vol. 103. No. 5-- 2025
Full Text