|
|
|
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
April 2026 | Vol. 104
No.8 |
|
Title: |
OPTIMIZING SEARCH SPACE FOR DYNAMIC SFC ROUTING: A MINIMUM SPANNING TREE-BASED
GREY WOLF ALGORITHM |
|
Author: |
ZAHIDA SHARIF , MUHAMMED BASHEER JASSER , ANGELA AMPHAWAN , KOK-LIM ALVIN YAU ,
TSE-KIAN NEO |
|
Abstract: |
In Network Function Virtualization Infrastructure (NFVI), the dynamic nature of
Service Function Chain (SFC) mapping poses significant challenges, particularly
for routing optimization. Fluctuating network conditions and user demands
rapidly expand the routing search space, making real-time identification of
optimal paths computationally expensive. An effective routing strategy therefore
requires the construction of an efficient search space that ensures full node
connectivity while selecting only qualitative links in terms of latency,
bandwidth, and resource availability. Achieving these objectives simultaneously
remains a major challenge, as conventional algorithms designed for rigid or
strong tree structures are often unsuitable for dynamic chain orchestration in
NFV environments. To address this issue, this work focuses on routing search
space optimization by reducing computational overhead and eliminating infeasible
routing options from the discrete solution space. A modified Grey Wolf
Optimization (GWO) algorithm is proposed, specifically tailored for discrete
routing scenarios in dynamic SFC mapping. The proposed approach employs a
discrete initialization strategy, enforces a Minimum Spanning Tree (MST) based
connectivity constraint to guarantee end-to-end reachability, and integrates a
penalty function to suppress redundant or overlapping routing solutions. This
design ensures that optimization is performed over a compact,
connectivity-aware, and quality-driven search space. Simulations results
validate the effectiveness of the proposed search space optimization strategy,
demonstrating significant performance improvements across different network
scales. The proposed approach achieves up to 91% reduction in execution time and
82.8% reduction in end-to-end delay, confirming its suitability for real-time
dynamic SFC routing. In addition, improved bandwidth efficiency is observed,
reflected by reductions of 51%,59%, and 66% in average available bandwidth for
small, medium, and large topologies, respectively. These quantitative gains
highlight the benefits of pre-optimizing the routing search space and confirm
its effectiveness for scalable and adaptive NFV environments. |
|
Keywords: |
Routing Optimization; Service Function Chaining (SFC); Search Space
Optimization; Swarm Intelligence; Latency Optimization |
|
DOI: |
|
|
Source: |
Journal of Theoretical and Applied Information Technology
30th April 2026 -- Vol. 104. No. 8-- 2026 |
|
Full
Text |
|
|
Title: |
MACHINE LEARNING CLASSIFICATION OF JUVENILE AND YOUTHFUL OFFENDERS USING
SOCIO-GEOGRAPHIC CRIME INDICATORS FROM MALAYSIAN ADMINISTRATIVE DATA |
|
Author: |
NURAZEAN MAAROP , GANTHAN NARAYANA SAMY , ROSLINA MOHAMMAD , NORZIHA MEGA MOHD
ZAINUDDIN , WAN ROSANISAH WAN MOHD |
|
Abstract: |
Examining the determinants of juvenile and youthful offending is essential for
understanding youth crime patterns and supporting evidence-based prevention
strategies. This study applies machine learning techniques to classify
individuals into juvenile offenders (children in conflict with the law) and
youthful offenders using administrative crime records obtained from Malaysia’s
Department of Social Welfare (Jabatan Kebajikan Masyarakat). The dataset
comprises 3,879 cases, including 2,743 children under the age of 18 and 1,136
youthful offenders aged between 18 and 21 years. The analysis incorporates
demographic attributes and socio-geographic crime indicators, including gender,
ethnic group, type of crime, state, state crime rate, state crime density,
population percentage, and crime ratio. Four supervised machine learning
algorithms, namely Logistic Regression, Decision Tree, Random Forest, and
Support Vector Machine, are employed to evaluate predictive performance. Model
evaluation is conducted using accuracy, precision, recall, and F1-score, with
stratified five-fold cross-validation applied to assess model robustness. The
results indicate that tree-based models outperform linear models in this
classification task. Decision Tree and Random Forest achieve the highest
performance, with an accuracy of approximately 97.6%, while Random Forest
attains a mean cross-validation accuracy of 97.86%. Feature importance and SHAP
analyses reveal that crime ratio, type of crime, state, and ethnic group are the
most influential predictors. Overall, the findings demonstrate that integrating
demographic characteristics with socio-geographic indicators enhances the
predictive classification of youth offenders. The proposed framework provides a
data-driven approach for analyzing administrative crime data and offers valuable
insights into regional and demographic patterns associated with youth offending. |
|
Keywords: |
Machine Learning, Crime Analytics, Juvenile Offenders, Random Forest,
Socio-Geographic Indicators, Data Mining. |
|
DOI: |
|
|
Source: |
Journal of Theoretical and Applied Information Technology
30th April 2026 -- Vol. 104. No. 8-- 2026 |
|
Full
Text |
|
|
Title: |
SOLVING DYNAMIC TASK OFFLOADING IN VEHICULAR FOG COMPUTING USING PARTIALLY
OBSERVABLE MARKOV DECISION PROCESS BASED DEEP DETERMINISTIC POLICY GRADIENT
APPROACH |
|
Author: |
J. SHANKAR , S. R. NAGARAJA |
|
Abstract: |
Vehicular Fog Computing (VFC) becomes the potential solution for mitigating the
vehicular computation load. In graded Vehicular Fog Computing (VFC), vehicles
function as mobile fog nodes at network edge, delivering consistent as well as
minimum-latency services. Due to limited onboard computational resources,
vehicles offload intensive tasks to nearby Roadside Units (RSU). To minimize a
computational weight at RSUs, VFC is utilized for the computational-exhaustive
tasks. Within this framework, vehicles serve as part of the infrastructure,
facilitating communication, monitoring, and resource sharing among fog nodes.
This makes efficient resource allocation a critical factor for overall system
performance. Thus, this research proposes the Partially Observable Markov
Decision Process-assisted Deep Deterministic Policy Gradient (POMDP-DDPG)
approach for an effective task offloading in VFC. The proposed approach is a
Reinforcement Learning (RL) that utilises the global data to identify a better
connection between the RSU and the fog servers. The experimental results specify
that the proposed POMDP-DDPG approach attains better results as compared to the
existing approaches |
|
Keywords: |
Deep Deterministic Policy Gradient, Partially Observable Markov Decision
Process, Reinforcement Learning, Roadside Units, Vehicular Fog Computing |
|
DOI: |
|
|
Source: |
Journal of Theoretical and Applied Information Technology
30th April 2026 -- Vol. 104. No. 8-- 2026 |
|
Full
Text |
|
|
Title: |
CONTRASTIVE SENTENCE EMBEDDINGS–BASED ZERO-SHOT DOCUMENT SUMMARIZATION USING
SINGLE-OBJECTIVE GENETIC ALGORITHM |
|
Author: |
KADRY HAMED, MOUSTAFA MAHMOUD AHMED ALI, YASER MAHER ABDELMONTELB, ABDELMGEID
AMIN ALI |
|
Abstract: |
Amid the rapid developments in the field of natural language processing (NLP),
few-shot and zero-shot text summarization have gained prominence as effective
methods for generating concise summaries from large texts. These methods employ
the capabilities of large language models (LLMs) to perform summarization tasks
with minimal or no specific training data. Despite the ability of these
approaches to generate open-domain, fluent, and coherent summaries, their
extractive summarization performance often remains lagging behind approaches
that leverage pre-trained language models (PLMs) with fine-tuning. Thus, we
propose a zero-shot extractive summarization framework that leverages the
capabilities of PLMs (without fine-tuning) combined with other unsupervised
learning strategies. This framework combines the open-domain generation
abilities of LLMs with the robust, high-precision performance typically
associated with fine-tuned PLMs. Recent advancements in attention mechanisms and
evolutionary algorithms have demonstrated significant potential in enhancing the
quality of document summarization. This work presents a novel extractive
open-domain summarization framework called ETS-BGA that integrates
attention-based neural networks with Genetic Algorithms (GAs) to generate
high-quality summaries, taking into account three summarization dimensions: (i).
Content relevance, which is obtained using a multi-head attention mechanism
applied to SimCSE (Simple Contrastive Learning of Sentence Embeddings) without
fine-tuning, (ii). Content coverage, which is fulfilled using a hybrid technique
that incorporates both Named Entity Recognition (NER) and TF-IDF-based keywords
extraction, and (iii). Redundancy minimization, which is obtained by computing
cosine similarity between sentence embeddings from SimCSE to catch deeper
semantic similarity beyond surface-level syntax. This work also presents a
comprehensive cross-paradigm comparative analysis covering statistical-based
methods, supervised deep learning-based methods, fine-tuned PLMs, and
zero/few-shot methods using LLMs. Experiments on CNN/DailyMail and the New York
Times benchmark datasets clarify that our proposed framework generates more
informative and less redundant summaries, achieving superior ROUGE scores
compared to many existing methods. These findings highlight the effectiveness of
combining contrastive learning with evolutionary optimization for zero-shot
summarization. |
|
Keywords: |
Extractive Text Summarization, Sentence Embeddings, Multi-Head Attention,
Zero-Shot Learning, Genetic Algorithms |
|
DOI: |
|
|
Source: |
Journal of Theoretical and Applied Information Technology
30th April 2026 -- Vol. 104. No. 8-- 2026 |
|
Full
Text |
|
|
Title: |
FASTER YOLO: AN EFFICIENT FRAMEWORK FOR CERVICAL CANCER CELL DETECTION WITH
DEFORMABLE CONVOLUTIONAL ATTENTION |
|
Author: |
JHEELAM MONDAL, RAJDEEP CHATTERJEE, MAHENDRA KUMAR GOURISARIA |
|
Abstract: |
Cervical cancer is the fourth most common disease worldwide. The most common
diagnostic method required for cervical cancer screening is the pap smear test.
Making precise diagnosis, identifying and classifying cells and closely
examining each slide all take a significant amount of time and work. Long
stretches of visual inspection can make human mistakes more likely thereby
resulting in incorrect classification of cells. An essential stage in automatic
cytopathology diagnosis is the detection of nuclei in cervical cell images. In
recent years, YOLO (You Only Look Once) models have been the most popular
paradigm in the field of real-time object detection because of their successful
balance between detection performance and processing cost. This work focuses on
a number of YOLO models and various cutting-edge object detection methods that
are trained on the popular SIPAKMED benchmark dataset. This dataset contains
annotated labels for each image. In this paper, we provide an improved
YOLO-based object detection model that achieves performance comparable to
state-of-the-art YOLO models while dramatically decreasing computing complexity.
The proposed model Faster YOLOv13s is built with an optimized attention aware
architecture that prioritizes efficiency above detection accuracy. Experimental
results show that the proposed model achieved a competitive mAP50 score of
87.00% compared to the best-performing YOLO model while significantly reducing
the number of trainable parameters and taking significantly less training time.
The results of this study are meant to guide future clinical applications and
identify the best model for cervical cancer detection. |
|
Keywords: |
Cervical Dysplasia, Image Pre-Processing, Object Detection, YOLO Network, Deep
Neural Network |
|
DOI: |
|
|
Source: |
Journal of Theoretical and Applied Information Technology
30th April 2026 -- Vol. 104. No. 8-- 2026 |
|
Full
Text |
|
|
Title: |
A QUANTUM CLASSICAL DEEP LEARNING FRAMEWORK FOR EARLY DETECTION AND TRIAGE OF
CARDIOVASCULAR DISEASES USING MEDICAL IMAGING |
|
Author: |
DR SELVANI DEEPTHI KAVILA, P. HEMA VENKATA RAMANA, SELVA MALAR N, CHILAKALAPUDI
MALATHI, DR DOKKU DURGA BHAVANI, DORABABU SUDARSA, N. SRIKANTH, MYLAVARAPU
KALYAN RAM |
|
Abstract: |
Early and accurate recognition of cardiovascular disease (CVD) using medical
imaging is essential for timely clinical intervention and management. The
objective of this study was to continue developing a hybrid quantum-classical
deep learning system for the early detection and automatic triage of
cardiovascular diseases using cardiac imaging information. The proposed method
consists of a combination of efficient feature extraction via a lightweight
classical convolutional neural network and a quantum-illness feature
transformation and decision-refinement process implemented as a variational
quantum circuit. A clinically informed triage module was added to rate patients
into three risk categories (low-, moderate-, and high-risk). Experimental
evaluation on a public cardiac magnetic resonance imaging (MRI) dataset revealed
that the proposed framework achieved 94.5% classification accuracy and an area
under the curve (AUC) of 0.96, with sensitivity exceeding that of
state-of-the-art classical deep learning models. The results show that
quantum-enhanced feature representations provide better class separability,
robustness to Imaging noise, and improved learning efficiency. The results
suggest that one way to overcome the limitations of conventional deep models is
the hybrid quantum-classical learning approach, which is also compatible with
near-term quantum hardware. This work raises the question of whether
quantum-assisted artificial intelligence can enable scalable, real-time
cardiovascular decision support systems and promote next-generation intelligent
healthcare systems. |
|
Keywords: |
Hybrid Quantum–Classical Learning; Cardiovascular Disease Detection; Medical
Image Analysis; Variational Quantum Circuits; Clinical Triage; Deep Learning |
|
DOI: |
|
|
Source: |
Journal of Theoretical and Applied Information Technology
30th April 2026 -- Vol. 104. No. 8-- 2026 |
|
Full
Text |
|
|
Title: |
AN AI-POWERED FRAMEWORK FOR REAL-TIME MALWARE ANALYSIS AND DETECTION IN CLOUD
ENVIRONMENTS TO ENHANCE SECURITY AND QOS |
|
Author: |
GAZALA BEGUM , DR. G. KRISHNA MOHAN |
|
Abstract: |
With the growing adoption of cloud computing, the challenge of dealing with
advanced cyberattacks, especially malware, becomes increasingly significant for
protecting data consistency and service reliability. In this paper, we present a
smart and scalable architecture for detecting malicious attacks in real-time
within the dynamic cloud environment. The architecture leverages a multi-tiered
approach to feature extraction that covers static, dynamic, and network-based
telemetry and processes this information using a combination of machine learning
algorithms such as Logistic Regression, Decision Trees, and Isolation Forest.
For added resilience and minimized false detections, the architecture features
an auto-recovery auto-activation scheme and an integrated cryptographic
architecture that safeguards individual nodes from malware threats. We validate
the efficacy of our design through experiments using standard malware data sets,
yielding an accuracy of 96.4% while keeping the false positive rate at 2.1%.
Moreover, our architecture improves resource efficiency by 18% under load
conditions due to intelligent workloads distribution. Through precision-based
threat detection and QoS-driven service delivery, our proposed solution ensures
adequate security in modern cloud infrastructure. |
|
Keywords: |
Cloud Security, Malware Detection, CNN-LSTM Architecture, Feature
Engineering, QoS-Aware Systems. |
|
DOI: |
|
|
Source: |
Journal of Theoretical and Applied Information Technology
30th April 2026 -- Vol. 104. No. 8-- 2026 |
|
Full
Text |
|
|
Title: |
A NOVEL HYBRID FRAMEWORK FOR BEHAVIOURAL PREDICTION AND INTERVENTION USING
ENHANCED EDUCATIONAL DATA MINING |
|
Author: |
V. PANDI SELVI, Dr. M. RAMASWAMI, M.SARAVANAKUMAR |
|
Abstract: |
The rapid growth of digital learning environments has generated vast amounts of
educational data, creating new opportunities for understanding student behavior
and improving learning outcomes. This study proposes a novel hybrid framework
for behavioural prediction and intervention using enhanced educational data
mining techniques. The framework integrates structural data analysis, machine
learning models, and pattern recognition methods to identify student learning
behaviours and predict potential academic risks.
The primary goal of this
research is to develop an intelligent system capable of early detection of
behavioural patterns such as disengagement, low performance, and irregular
participation. By combining multiple analytical approaches, the proposed model
enhances prediction accuracy and supports timely, personalized interventions.
The hypothesis of this study is that a hybrid data mining approach will
significantly outperform traditional single-model methods in predicting student
behaviour and improving intervention effectiveness.
Experimental results
demonstrate that the proposed framework provides improved prediction reliability
and supports educators in making data-driven decisions. This research
contributes to the advancement of educational analytics by offering a scalable
and adaptive solution for modern learning environments. |
|
Keywords: |
Academic Performance, Association Rule Mining, Behavioural Prediction, Bayesian
Networks, Educational Data Mining, Hybrid Framework, Machine Learning, Real-Time
Intervention, Student Behaviour |
|
DOI: |
|
|
Source: |
Journal of Theoretical and Applied Information Technology
30th April 2026 -- Vol. 104. No. 8-- 2026 |
|
Full
Text |
|
|
Title: |
A DATA-DRIVEN APPROACH FOR EQUIPMENT FAILURE PREDICTION IN DATACENTRE NETWORKS |
|
Author: |
TAIWO AMOO, VICTOR ODUMUYIWA, OLADIPUPO SENNAIKE |
|
Abstract: |
A datacenter network comprises intermediary devices such as servers, routers,
switches, etc., which are interconnected to allow for seamless exchange of
information. These intermediary devices in the network are configured to record
events occurring at any point in time, thereby depicting the current condition
of these devices. However, these event logs, being massive datasets, are
characterized by high volume, velocity and variety, making the manual-based
analysis difficult. Recently, machine learning algorithms have been proposed in
predicting failures using time-based sequential data such as log data, with
Support Vector Machine (SVM) achieving remarkable success. Despite the
robustness and high accuracy of SVM, it is not suitable for classifying large
datasets due to the complexity of training relative to the volume of data and
the high cost of manually labelling large datasets. To this end, a recent study
employed SVM with a weakly supervised learning approach called Multi Instance
Learning (MIL) to reduce the labelling effort required for large datasets in
classification tasks. However, these MIL-SVM-based algorithms do not account for
temporal information when addressing classification tasks. They also fail to
account for the consistent shift in data distribution associated with big data,
particularly in event logs, when selecting SVM hyperparameters during training
for classification tasks. This work therefore proposed a Multi-Instance
Selector-SVM (MIS-SVM) based on a multi-objective hyper-heuristics framework for
failure prediction in a datacenter network. The Multi-Instance Selector (MIS)
algorithm was developed to select likely positive and negative instance
representatives per group of instances (bag) to reduce noise in both positive
and negative bags. Subsequently, a hyper-heuristic framework using the
lexicographic multi-objective genetic algorithm was developed to select
competitive MIS-SVM hyperparameter values to improve generalization performance
on time-based sequence data, such as logs. The proposed model was evaluated
using three conflicting objectives —False Positive Rate (FPR), False Negative
Rate (FNR), and Number of Support Vectors (NSV)—on three publicly available log
datasets. The results showed that the proposed Hyper-heuristic-based MIS-SVM
yielded the maximum p-value of 0.008 on M1 dataset, 0.001 on M2 dataset, and
0.005 on M3 dataset, with a p-value of 0.05 as the significance level. This
demonstrated that the hyper-heuristic-based MIS-SVM is statistically superior to
existing methods in the literature. |
|
Keywords: |
Machine Learning; Event Logs; Hyper-parameter Optimization; Multi-Instance
Learning |
|
DOI: |
|
|
Source: |
Journal of Theoretical and Applied Information Technology
30th April 2026 -- Vol. 104. No. 8-- 2026 |
|
Full
Text |
|
|
Title: |
ENHANCING CLOUD SECURITY IN INDUSTRY 4.0: A ROBUST IDENTITY AUTHENTICATION
SYSTEM |
|
Author: |
USHA V, SRIDHAR M |
|
Abstract: |
Solid cloud security could be a squeezing concern within the setting of Industry
4.0, which is characterized by the meeting of advanced and fabricating
innovations. The need of dependable personality identification procedures in
industrial transmission environments is discussed in this study as a significant
barrier to progress in the manufacturing sector. Currently, there is no identity
authentication in the system's architecture. Here, the omnipresent data
collectors that make up the IoT send their data to a backend server via the
time-tested Modbus protocol. Carefully designed to strengthen data security in
the complex fabric of industrial settings, this identity identification system
is a labour of love. Two main servers, the Register Server (RS) both the Backend
Data Analysis Server (BDAS), the foundation of this system. A trusted authority,
RS not only verifies the authenticity of IoT terminal devices but also registers
their identities and generates JWT tokens. Beginning their trip with a
registration request to RS, terminal IoT devices are then issued their own
temporary identifiers (TIDi) by RS. Concurrently, RS produces tokens for
effective devices, acting as a portal for encrypted data transfer. The
methodology of the system breaks down the procedure into four discrete parts,
including identity registration, first-stage authentication, token acquisition,
and data transmission, each carefully divided to meet the highlighted
difficulties. There are also three distinct phases to the authentication
procedure: setup/initialization, identity registration, and authentication. In
the first phase of identity registration, the registration server disseminates
crucial information. There are five stages to the identity registration process,
beginning with a registration request from the terminal IoT device and ending
with information sharing and the coordination of individual TIDi identities. The
identity identification phase is the most crucial, and it requires a
comprehensive 13-step verification process that is broken down in detail from
the point of view of various devices. As a whole, this article proposes a
well-thought-out identity authentication methodology to strengthen cloud
security in the Industry 4.0 scenario, providing a methodical means of dealing
with the problems that have been highlighted. |
|
Keywords: |
Industry 4.0, Cloud security, Manufacturing sector, Cryptography, Security |
|
DOI: |
|
|
Source: |
Journal of Theoretical and Applied Information Technology
30th April 2026 -- Vol. 104. No. 8-- 2026 |
|
Full
Text |
|
|
Title: |
METHODOLOGICAL FOUNDATIONS OF AI-DRIVEN DISTANCE LEARNING PLATFORMS: EVIDENCE
FROM LEARNING ANALYTICS AND STRUCTURAL
EQUATION MODELING |
|
Author: |
MAXOT RAKHMETOV , AINUR SISSENOVA , RAUSHAN MOLDASHEVA , AKMARAL DAUTKULOVA,
KAMKA UTEULIYEVA, BALAUSSA ORAZBAYEVA |
|
Abstract: |
In the context of active digitalization of education, this study examines the
effectiveness of distance learning in a digital educational environment based on
the application of artificial intelligence principles in the design of
educational platforms. The purpose of the research is to develop and empirically
test a methodology for creating a distance learning platform that provides
adaptive learning trajectories, personalization of educational content, and
pedagogical decision-making based on data analysis.
As part of the research,
a structural and functional model of an intelligent educational platform was
developed and its pilot implementation in higher education was carried out.
Empirical data was collected from students with different levels of digital
competence and training. To assess the effectiveness of the proposed
methodology, the methods of learning analytics, comparative analysis of
educational results, as well as statistical modeling of educational data were
used.
The results of the study show that adaptability and personalized
feedback, implemented using artificial intelligence methods, are key factors in
improving learning efficiency at the initial stages of using the platform, while
student analytics and predicting learning outcomes become increasingly important
at subsequent stages of learning. It has been established that continuous
modeling of students' educational behavior and automatic correction of
educational content contribute to increased engagement and sustained growth in
academic performance.
In addition, differences in the effectiveness of
learning between groups of students with different levels of initial digital
readiness have been identified, which con-firms its moderating role in distance
learning. The results obtained emphasize the need to integrate the principles of
artificial intelligence into the methodological foundations of the development
of distance learning platforms in order to increase personalization, optimize
cognitive load and improve educational outcomes information |
|
Keywords: |
Artificial Intelligence; Distance Learning; Digital Educational Platforms;
Adaptive Learning; Personalized Learning; Learning Analytics; Digital Readiness;
Learning Effectiveness; Structural Equation Modeling; Higher Education. |
|
DOI: |
|
|
Source: |
Journal of Theoretical and Applied Information Technology
30th April 2026 -- Vol. 104. No. 8-- 2026 |
|
Full
Text |
|
|
Title: |
CRYPTOSYSTEM USING RNA COMBINED WITH DATA SEQUENCE AND AFFINE CIPHER FOR SECURE
DATA COMMUNICATION |
|
Author: |
KAMERAN ALI AMEEN, ASO AHMED MAJEED, YALMAZ NAJM ALDEEN TAHER, YOUSIF MOHAMMED
WAHAB, KAWA M KAKY |
|
Abstract: |
The 21st century has experienced an information surge due to rapid technological
advancement, rendering knowledge a much more vital strategic asset, particularly
when transmitted across insecure means and vulnerable to intrusions.
Cryptosystems are commonly employed methods for safeguarding user data from
undesirable access over communication networks. Moreover, it is essential to
devise cryptographic solutions that are equilibrated in terms of security and
efficiency. Bio cryptography is a new and promising cryptographic research area.
Besides, Ribonucleic Acid (RNA) cryptography has exhibited considerable
effectiveness. Moreover, research indicates that integrating the Affine Cipher
with RNA sequence-inspired encoding, which depends on data sequences, is
effective. The proposed cryptosystem utilizes an Affine Cipher initially to
encrypt the original message through several procedures. The output is
subsequently encoded into data sequences by generating a six-bit sequence.
Ultimately, it encodes it according to the fundamental Biological RNA sequence.
Comparative studies have been performed to validate the efficacy of our concept.
The proposed method met the security requirements and showed the capability to
counter many security threats. The results indicate that our method surpasses
its alternative in terms of time and complexity, thereby meeting the majority of
security objectives while ensuring high levels of privacy, security, and a
respectable IC. It is also robust against certain recognized attacks. Moreover,
the proposed algorithm yields faster execution times and enhanced performance
relative to other algorithms during the encryption and decryption of 24B and 27B
texts. |
|
Keywords: |
Affine Cipher, XOR operation, Data Sequence, RNA Sequence, Cryptography,
Security. |
|
DOI: |
|
|
Source: |
Journal of Theoretical and Applied Information Technology
30th April 2026 -- Vol. 104. No. 8-- 2026 |
|
Full
Text |
|
|
Title: |
ASSESSING TECHNOLOGY READINESS FOR BLENDED LEARNING AMONG STUDENTS IN INDONESIAN
ISLAMIC HIGHER EDUCATION INSTITUTIONS |
|
Author: |
MEINARINI CATUR UTAMI , EVY NURMIATI , CAHYO CRYSDIAN , QURROTUL AINI |
|
Abstract: |
Since the pandemic, higher education has further embraced digital
transformation, with blended learning being viewed as a viable and pertinent
strategy. The digital gap and students' inclination for in-person instruction
are just two of the difficulties that Islamic Higher Education Institutions
(IHEIs) in Indonesia encounter while implementing blended learning, despite the
growing advancement of technology infrastructure. By combining the Six Learning
Aspects with the Readiness for Blended Learning approach, this study seeks to
assess Java IHEIs students' readiness for blended learning. Purposive sampling
methods were used to gather data from 348 respondents at Java IHEIs. PLS-SEM was
used for analysis with SmartPLS 3.3.3. The findings indicate that while
technological accessibility has no discernible impact on blended learning
preparation, attitudes about classroom instruction are the most important
predictor. This result demonstrates that affective preferences and learning
culture influence IHEIs students' preparedness for blended learning, alongside
technological considerations. The study's findings highlight the crucial role of
human-centered approaches in IHEIs 's development of blended learning systems.
Thus, while continuing to improve faculty competency and technology
infrastructure, institutional policies should prioritize the development of
learning flexibility, consistent individual study habits, and
motivation-boosting initiatives. |
|
Keywords: |
Digital Gap; SEM; Blended Learning; Readiness; Six Learning Aspects; Readiness
for Blended Learning |
|
DOI: |
|
|
Source: |
Journal of Theoretical and Applied Information Technology
30th April 2026 -- Vol. 104. No. 8-- 2026 |
|
Full
Text |
|
|
Title: |
PERFORMANCE ANALYSIS OF LOG-TYPE LIFETIME DISTRIBUTION BASED ON INFINITE FAILURE
NHPP SOFTWARE RELIABILITY MODEL |
|
Author: |
HYO JEONG BAE |
|
Abstract: |
This study analyzes the performance of an infinite-failure NHPP software
reliability model based on logarithmic time transformation by applying a
Log-Type lifetime distribution, which is effective for complex reliability
analysis as it can represent various types of failure occurrences. Software
failure time data are employed, and the model parameters are estimated using the
MLE approach. The proposed models are evaluated using multiple criteria,
including goodness-of-fit measures such as MSE and R², predictive performance
assessed via m(t), failure occurrence intensity characterized by λ(t), and
reliability measured by R̂(τ). The analysis results indicate that the
Log-Poisson model outperforms the competing models. Consequently, this study
systematically extends and validates the reliability performance of Log-Type
distribution families that have not been sufficiently examined in previous
studies, and the findings are expected to provide useful baseline information
for early-stage software failure rate analysis. |
|
Keywords: |
Infinite-failure, Log-Linear, Log-Poisson, Log-Power, Log-Type, NHPP. |
|
DOI: |
|
|
Source: |
Journal of Theoretical and Applied Information Technology
30th April 2026 -- Vol. 104. No. 8-- 2026 |
|
Full
Text |
|
|
Title: |
EXPLAINABLE AI FOR CREDIT RISK MANAGEMENT IN REGULATED FINANCIAL ENVIRONMENTS: A
TRANSPARENT AND AUDITABLE DECISION FRAMEWORK |
|
Author: |
SUNILKUMAR REDDY ERAGANENI, VENKATA SUDHAKARA REDDY A, KOTI REDDY UPPALA |
|
Abstract: |
As artificial intelligence becomes popular for credit risk, the issues of
transparency, explainability, and regulatory compliance raise significant
challenges. Even though sophisticated predictive techniques can improve risk
assessment accuracy, the black-box nature of the decision process associated
with those techniques is frequently at odds with the regulated financial
industry's reliance on explainability, accountability, and auditability. This
study introduces an explainable artificial intelligence (XAI)–driven framework
for credit risk management to support transparent, regulator-aligned
decision-making. The proposed method provides both global and local
interpretability, allowing stakeholders to reason about how important financial
and credit factors drive risk at both levels. On the model side, explainability
techniques integrated into the model itself yield human-interpretable
explanations for model predictions, which ultimately support internal risk
governance, compliance audits, and fair lending practices. It provides an
explainable decision logic framework that meets regulatory expectations for
model transparency while helping avoid the risks of poorly deploying a black-box
AI model. These results highlight that risk assessment methods, while remaining
consistent with regulatory principles, can provide reliable and consistent
credit risk insights when they also enforce explainability guidelines. The
framework provides traceable, interpretable, and defensible credit decisions
that instil trust among financial institutions, regulators, and customers. To
sum up, this work improves upon the existing state of the art in accountable
financial analytics by exemplifying an explainability-based paradigm of credit
risk. This sets the stage for trust and opens new conversations that
explainability itself is a foundational component of both compliance and AI
solution trust in regulated financial institutions. |
|
Keywords: |
Explainable Artificial Intelligence (XAI), Credit Risk Assessment Model,
Transparency and Interpretability Regulatory, Compliance in Financial Systems,
Auditable Decision-Making Framework |
|
DOI: |
|
|
Source: |
Journal of Theoretical and Applied Information Technology
30th April 2026 -- Vol. 104. No. 8-- 2026 |
|
Full
Text |
|
|
Title: |
DIGITAL TECHNOLOGIES AS A TOOL FOR ADAPTIVE TEACHING OF FOREIGN LANGUAGES IN
HIGHER SCHOOL BASED ON ARTIFICIAL INTELLIGENCE |
|
Author: |
OLHA TAMARKINA, LIUDMYLA BAIDAK, HANNA TSYHANOK, NATALIIA SHCHUR, ANNA ZINCHENKO |
|
Abstract: |
Current digital transformation of education necessitates the creation of
intelligent solutions capable of ensuring personalized and effective learning of
foreign languages in higher school. The relevance of the study is determined by
the need to introduce adaptive digital technologies based on artificial
intelligence (AI) into the teaching of foreign languages in higher school to
personalize learning and increase its efficiency. The aim of the study is to
clarify the effectiveness of the use of AI-based digital technologies as a tool
for adaptive teaching of foreign languages in higher education institutions
(HEIs). The object of analysis was the architecture of adaptive educational
systems that integrate machine learning (ML), deep neural networks (DNNs), deep
learning (DL), natural language processing (NLP), and intelligent analytics
algorithms to form dynamic learning trajectories. The methodology included
modelling adaptive learning scenarios in Moodle and OpenEdX environments using
AI plugins, evaluation by the integral IAAL index (Acc, Stab, Latency, CogFit),
as well as statistical methods (analysis of variance (ANOVA), bootstrapping,
receiver operating characteristic (ROC)/ area under the curve (AUC), principal
component analysis (PCA), k-means). The international perspective was provided
by analysing the practices of using digital adaptive learning algorithms (ML,
DL, NLP modules) in three countries — Ukraine, Germany, and Poland. The analysis
identified differences in the performance of models in environments with
different levels of digital maturity. The results showed that the hybrid
Fusion-NLP configuration achieved the highest integral Adaptive Academic
Learning Index (AALI) values (0.82–0.87), ensuring stability and cognitive
relevance of learning tasks. Deep-Adapt demonstrated maximum accuracy in Germany
(Acc = 0.89; AUC = 0.93), while Stat-Learn confirmed its superiority in speed in
Poland and Germany (Latency = 0.92–0.93). It was found that the effectiveness of
the models depends on the educational context. Fusion-NLP is the most effective
among the environments with fragmented data. Deep-Adapt is the most effective
among mature systems with complete data sets. Stat-Learn is optimal for rapid
diagnostics in transitional infrastructures. The academic novelty is the
comprehensive comparison of the three architectures using the AALI and
multidimensional methods, which allowed us to determine the optimal application
scenarios in different countries. Further research prospects include improving
AALI through cognitive indicators, expanding the international sample, and
developing AI-based monitoring dashboards for teachers, which will facilitate
the practical integration of adaptive technologies into language teaching. |
|
Keywords: |
Artificial Intelligence, Digital Technologies, Adaptive Learning, NLP, Machine
Learning, Higher Education, International Perspective |
|
DOI: |
|
|
Source: |
Journal of Theoretical and Applied Information Technology
30th April 2026 -- Vol. 104. No. 8-- 2026 |
|
Full
Text |
|
|
Title: |
DEVELOPMENT OF A CONCEPTUAL FRAMEWORK FOR AN EXPLAINABLE ARTIFICIAL
INTELLIGENCE-BASED PREDICTIVE MAINTENANCE DECISION SUPPORT SYSTEM FOR THE
TRAFFIC ELECTRICAL INDUSTRY |
|
Author: |
YOTSAPORN PUGDEECHON, SUWUT TUMTHONG |
|
Abstract: |
This research aims to develop a conceptual framework for an explainable
artificial intelligence-based predictive maintenance decision support system
(XAI-PdM DSS) for the electric traffic industry. This system uses specialized
machinery that is susceptible to deterioration and unexpected downtime. The
research addresses the limitations of traditional maintenance and black-box AI
systems, which lack transparency in decision-making. This developmental research
was conducted through a literature review on decision support systems,
predictive maintenance, machinery life expectancy forecasting, explainable
artificial intelligence, and multi-criteria decision techniques. This review
synthesized and designed a conceptual framework that integrates AI models for
machinery deterioration forecasting with machinery deterioration prediction
modeling techniques to explain forecasting results. Multi-criteria decision
techniques are used to prioritize maintenance tasks, presenting the results
through a dashboard to support systemic decision-making. An expert evaluation of
the conceptual framework by 15 individuals found it highly suitable (𝑥̄ = 4.23,
S.D. = 0.58), indicating that the developed framework has the potential to
support efficient, transparent, and practical predictive maintenance
decision-making in the electric traffic industry. |
|
Keywords: |
Decision Support System; Predictive Maintenance; Explainable Artificial
Intelligence; Remaining Useful Life. |
|
DOI: |
|
|
Source: |
Journal of Theoretical and Applied Information Technology
30th April 2026 -- Vol. 104. No. 8-- 2026 |
|
Full
Text |
|
|
Title: |
a - D NN: OPTIMIZED-ATTENTION GUIDED DEEP NETWORK FOR PREDICTION OF CARDIAC
ARREST |
|
Author: |
VINITHA V , PROF. DR. V. PARTHASARATHY , DR.R. SANTHOSH |
|
Abstract: |
The cardiologist can attend most cardiac arrests successfully by evaluating the
various risk factors of patients. The ECG signals provide quantitative
prognostics information with constant monitoring without considering random
temporal information. This research model a novel Attention- Deep Neural Network
model (𝑎 − 𝐷NN) with an optimization approach to capture the essential
information for the earlier prediction of the neurological outcomes which causes
sudden cardiac arrest. The proposed model includes three diverse phases: 1)
Min-Max data normalization; b) Bounding-based weighted normalized feature
learning; and c) prediction. Initially, the input is taken from the available
online dataset for cardiac arrest. The dataset is verified for the minimal and
maximal level of attribute range for a certain level. The bounding range of the
data is provided with a weighted value for learning the features. The weighted
function is multiplied with the dataset attribute values to measure the
deviation level of the heart rate. The dynamic weighted function is optimized
with stochastic Golden Eagle Optimizer (GEO) using Bellman Equation (BE). The
optimal heart rate features are subjected to the Deep network model. This
hybridized learning model is optimized with weights considering weights using an
appropriate prediction algorithm. Furthermore, the proposed (𝑎 − 𝐷NN -GEO)
comparative analysis is done with various existing approaches like Random Forest
(RF), Decision tree (DT), Long Short-Term Memory (LSTM), Logistic Regression
(LR) and hybrid approaches. Different performance metrics like sensitivity,
specificity, PPV, NPV, and F1-score are evaluated and compared. |
|
Keywords: |
Cardiac Arrest; Deep Learning; Golden Eagle Optimizer; Deep Belief Network;
Normalization |
|
DOI: |
|
|
Source: |
Journal of Theoretical and Applied Information Technology
30th April 2026 -- Vol. 104. No. 8-- 2026 |
|
Full
Text |
|
|
Title: |
LINEAR MULTIPLICATIVE CRYPTOSYSTEM BASED DEEP NEURAL LEARNING CLASSIFIER MODEL
FOR DATA TRANSACTION SECURITY IN CLOUD COMPUTING |
|
Author: |
SHAKIRA P V , LAXMI RAJA |
|
Abstract: |
Cloud computing has emerged as a significant platform due to its rapid condition
of computing resources for data analysis. Cloud data are stored and accessed on
remote servers, allowing users to access them from anywhere at any time. As
number of CU increases, there is a growing need to protect the data of various
users. Security is main concern because information is broadcasted to remote
servers more than channel. Previous to data transactions occur at cloud
computing, safety demands required to mentioned. Therefore, a novel Linear
Multiplicative Cryptosystem-based Deep Neural Learning Classifier (LMC-DNLC)
model has been developed to facilitate secure data transactions from cloud users
to cloud servers with enhanced data confidentiality and reduced time
consumption. The LMC-DNLC model encompasses four processes namely key
generation, encryption, classification, and decryption. Each user registers
their details and generates session key pairs using the Joye-Libert Cryptosystem
with the assistance of a Linear Multiplicative generator. Subsequently, user
data is encrypted using Joye-Libert encryption and transmitted to CS. On server
side, the classification of authentic or illegitimate users is performed by
DNLC. The Forbes similarity coefficient is applied to deep learning for
analyzing user keys, classifying genuine users with the help of a soft-step
activation function. Finally, genuine users are decrypted using Joye-Libert
decryption to attain original information. This process ensures secure data
transactions with higher data confidentiality between users and cloud servers.
Experimental assessment is conducted on various factors concerning number of CU
and data. Outcomes of quantitative analysis indicate that the LMC-DNLC model
provides an efficient solution for attaining superior data confidentiality and
integrity, reducing communication, computation overhead compared to existing
methods. |
|
Keywords: |
Cloud, Security In Data Transaction, Deep Neural Learning Classifier,
Joye-Libert Cryptosystem, Linear Multiplicative Generator, Forbes Similarity
Coefficient, Soft-Step Activation Function, |
|
DOI: |
|
|
Source: |
Journal of Theoretical and Applied Information Technology
30th April 2026 -- Vol. 104. No. 8-- 2026 |
|
Full
Text |
|
|
Title: |
CHSEGNET: COMPOSITE HEART SEGMENTATION NETWORK FOR CARDIAC IMAGE SEGMENTATION |
|
Author: |
NANDHAGOPAL SUBARAMANI , E SASIKALA |
|
Abstract: |
Cardiac MRI is vital in following the disease trajectory and treatment response.
By providing detailed, quantitative data about the anatomy and function of the
heart, cardiac MRI allows clinicians to accurately measure cardiac function is
an invaluable resource for identifying disease progression and response to
therapy. Segmentation employing DL allowed us to accurately contour the
myocardial, right ventricle and left ventricle. The detailed cutting not only
supports the diagnosis of heart failure, but also the understanding of the
condition of the heart's operations following therapy and the response to
therapies. Our objective is to create a Composite Heart Segmentation Network
(CHSegNet) that leverages MRI imaging data to segment cardiac organs. The
complex architecture of the heart with its many chambers, arteries, and tissues
makes it difficult to achieve high segmentation effectiveness. Furthermore, the
heart itself beats, inducing motion artifacts that degrade the quality and
reliability of images. Consequently, a CHSegNet approach is investigated to
address several issues in cardiac MRI image segmentation. A CHSegNet system is
proposed. It’s built upon the encoder-decoder pair that boosts cardiac semantic
segmentation effectiveness with the dense layer. To maintain a spatial
information, information from heavily convolutional regions of various sizes is
combined by the dense level. The densely connected layer, enables the network to
deal with variables at various spatial scales to record the worldwide as well as
local context. This retains the cardiac structural detail and context required
for high-fidelity segmentation of the larger cardiac architectures. To enable
the deep connection to learn important semantic variables, the encoder's higher
level receives the dense layer. With the roll of an intersection over union
(IoU), and precision scores of 78.1%, 97.9% and accuracy of 99.5%
correspondingly, the encoder layer generated the best effectiveness for
segmentation according to an investigation of several situations and structural
modifications. The assignment of a shallower stratum like more geographic
information is preserved by encoder layer necessitating greater parallelism
routes to capture the ideal set of variables since different configurations have
varied effects on the network. Conversely, deeper layers, have fewer parallel
routes needed to provide the best network because they have more useful
characteristics but a lower spatial resolution. |
|
Keywords: |
Cardiac Segmentation, Deep Learning, Prediction, Dense Layer, Accuracy |
|
DOI: |
|
|
Source: |
Journal of Theoretical and Applied Information Technology
30th April 2026 -- Vol. 104. No. 8-- 2026 |
|
Full
Text |
|
|
Title: |
COMPREHENSIVE SKIN DISEASE CLASSIFICATION: ENHANCING EARLY DETECTION THROUGH
ADVANCED IMAGE PROCESSING AND MACHINE LEARNING MODELS |
|
Author: |
JESWIN PAUL PANDIAN , SAMSON ISAAC , P. SUBHA HENCY JOSE , P. ANANTHA CHRISTU
RAJ |
|
Abstract: |
Skin disease varies from benign blemishes to cancerous malignancies. This study
uses large datasets to classify skin diseases as malignant or non-cancerous to
improve early diagnosis and treatment. Our study examines basal cell carcinoma,
squamous cell carcinoma, actinic keratosis, seborrheic keratosis, melanoma,
granulomas, angiomas, and angiokeratomas, as well as benign conditions like
melanocytic nevus, dermatofibroma, hemangioma, pilar cysts, keratinous cysts,
moles, skin tags, and wart for comparison, healthy skin photos are supplied. Two
datasets—10,000 melanoma photos and 5,000 images of various skin
conditions—represent varied skin types and lesion appearances. Preprocessing
included scaling all photos to uniform dimensions, image augmentation (flipping,
rotating, scaling), and pixel value normalization to aid machine learning
algorithm convergence. Dermatologists cleaned data and tagged photos with
condition-specific information for supervised learning. We oversampled minority
classes and created synthetic data using SMOTE to correct class imbalance. Our
study focuses on segmentation and feature extraction to properly identify skin
cancer, non-cancerous lesions, and healthy skin. For exact lesion border
delineation, MorphACWE, MorphGAC, thresholding, edge detection, clustering, and
semantic segmentation were used. Both standard and sophisticated approaches were
used to extract features, including HOG, LBP, GLCM, Gabor filters, and shape
descriptors like Hu and Zernike moments. Skin disease classification was tested
using five machine learning models: ResNet/DenseNet, GAN, Random Forest (RF),
Support Vector Machine (SVM), and K-Nearest Neighbor (KNN). Our dataset
contained 10,000 melanoma photos, 10,000 healthy skin images, and images of
actinic keratosis, melanocytic nevus, dermatofibroma, hemangioma, keratinous
cysts, pilar cysts, lipoma, moles, skin tags, and warts. We choose the best skin
disease classification model by comparing F1-score, recall, accuracy, and
precision. This study emphasizes the need to combine image processing
technologies to improve skin lesion identification. We want to improve
dermatology by allowing quick and exact detection of skin problems, which will
improve clinical choices and treatment results. The report also recommends
dermatological diagnostics research and examines model selection in practical
problems healthcare settings. |
|
Keywords: |
Skin Diseases, Melanoma, Image Segmentation, Feature Extraction, Machine
Learning, Classification, Dermatology. |
|
DOI: |
|
|
Source: |
Journal of Theoretical and Applied Information Technology
30th April 2026 -- Vol. 104. No. 8-- 2026 |
|
Full
Text |
|
|
Title: |
INNOVATIVE AI TOOLS FOR DIGITAL BUSINESS TRANSFORMATION WITH A FOCUS ON DATA
MANAGEMENT AND AUTOMATION OF KEY PROCESSE |
|
Author: |
KATERYNA HALAN, NATALIIA HAVADZYN , TETIANA KNIAZIEVA , ELENA LYTVYNENKO,
NATALIA DANIK |
|
Abstract: |
This study aims to determine whether companies’ management of data and AI-driven
automation contributes to operational flexibility and competitiveness, focusing
on highly specialized businesses. The authors employ a global perspective using
expert judgment, cross-country comparisons, compliance with international
standards (ISO/IEC 27001:2022, ISO/IEC 38505-1:2021, ISO/IEC 42001:2023), and
combined quantitative tools within the "Digital Transformation Integration
Index" (DTII) model. Empirical research focuses on the UAE, Ukraine, and
Singapore, each with distinct regulatory regimes, technical infrastructures, and
organizational maturity levels. Digital capability is assessed through three
factors: data governance capacity (DGC), AI automation performance (AAP), and
organizational readiness for AI adoption (ORA). Results establish a hierarchy of
digital maturity. Singapore achieves the highest DTII score (.94) due to robust
data governance, advanced automation, and strong organizational capabilities.
The UAE scores second (.815), benefiting from rapid public-sector modernization
and fast AI implementation, presenting significant scaling opportunities.
Ukraine’s DTII score (.584) reflects improving digital skills and automation but
highlights the need for enhanced infrastructure and regulation, described
through a “transitive growth” model. Sectoral AAP analysis shows Singapore
leading (88–95), the UAE with high but uneven performance (79–90), and Ukraine
at average levels (63–78). Differences are also observed in cultural orientation
toward innovation, managerial digital literacy, and institutional readiness for
AI adoption, as measured by ORA. The study provides a comprehensive analytical
framework to evaluate a company’s or country’s preparedness for digital
transformation and to guide strategic priorities. Its findings can support
corporate digital strategy development, enhance data management systems,
streamline AI-driven workflows, and inform policies to stimulate digital economy
growth, particularly in developing countries. |
|
Keywords: |
Digital Transformation, Artificial. Intelligence, Information Governance,
Process. Automation, Planning Readiness, AI Integration, Business. Analytics,
Digital Strategy, Innovation, Systems, Sustainable. Development. |
|
DOI: |
|
|
Source: |
Journal of Theoretical and Applied Information Technology
30th April 2026 -- Vol. 104. No. 8-- 2026 |
|
Full
Text |
|
|
Title: |
HYBRID PHYSICS-INFORMED NEURAL NETWORK AND EXTENDED KALMAN FILTER FOR ACCURATE
STATE-OF-HEALTH ESTIMATION OF BATTERIES IN ELECTRIC VEHICLES |
|
Author: |
VIJAYA ANAND NIDUMOLU , HEMANTH SAI MADUPU , PADALA SRIKAVITHA , VENKATA ASHOK
KALAGA , NIMMAGADDA CHANDRA SEKHAR |
|
Abstract: |
Accurate battery health monitoring is critical for the reliable operation of
electric vehicles (EVs). Existing approaches for State-of-Health (SOH)
estimation often face challenges such as high computational cost, reliance on
large pre-training datasets, sensitivity to irregular operational data, and
limited generalizability under diverse real-world conditions. To address these
limitations, this work proposes a hybrid framework that integrates a
Physics-Informed Neural Network (PINN) with an Extended Kalman Filter (EKF) for
real-time, robust SOH estimation. The PINN captures nonlinear relationships
between battery operating parameters and SOH while enforcing physical
degradation constraints, whereas the EKF enables recursive prediction and
correction, mitigating the effects of measurement noise and modeling
uncertainties. The proposed PINN–EKF framework achieves accurate, stable, and
physically consistent SOH predictions under varying operating conditions.
Performance evaluation along with comparative analysis against existing methods,
demonstrates that the proposed approach provides superior estimation accuracy,
computational efficiency, and robustness, making it highly suitable for
practical deployment in resource-constrained EV systems. |
|
Keywords: |
State of Health, Extended Kalman Filter, Battery Management, Lithium Ion
Battery, Useful Life Estimation. |
|
DOI: |
|
|
Source: |
Journal of Theoretical and Applied Information Technology
30th April 2026 -- Vol. 104. No. 8-- 2026 |
|
Full
Text |
|
|
Title: |
DATA SECURITY IN DISTRIBUTED DATABASES: MODERN ENCRYPTION AND AUTHENTICATION
METHODS |
|
Author: |
SERGII BATAIEV , VIKTOR KYRYCHENKO , VOLODYMYR STANKO , SERHII VOLOSHCHUK ,
TARAS STARUSHENKO |
|
Abstract: |
Distributed databases are a core component of modern information systems, and
their security has become increasingly critical. According to the EU Agency for
Network and Information Security (ENISA) Threat Landscape 2024, cyberattacks on
distributed systems increased by 47% over the past year, while traditional
security mechanisms designed for centralized architectures fail to address
vulnerabilities related to inter-node communication, data replication, and
consensus protocols. This study aims to develop a universal approach to securing
distributed databases by systematizing architecture-specific threats, conducting
a comparative analysis of contemporary encryption and authentication mechanisms,
and proposing an architectural security framework tailored to sharding-based,
master–slave replication, and consensus-based systems. Threats were classified
along three dimensions—architectural level, compromise type, and attack
vector—using a systematic mapping study. Multi-Criteria Decision Analysis (MCDA)
was applied to evaluate protection methods by security, performance,
implementation complexity, and compliance with ISO/IEC 27001:2022 and PCI DSS
v4.0.1, and to construct a decision tree for selecting appropriate security
controls. The analysis draws on 87 peer-reviewed publications (Scopus, IEEE,
ACM; 2020–2025), documentation of Apache Cassandra, MongoDB, and CockroachDB,
official NIST and ENISA reports, and empirical data from 245 CVE
vulnerabilities. Emerging threats for 2024–2025 were identified using MCDA
weighting, Spearman correlation analysis (n = 87, p < 0.05), and YAKE keyword
extraction. Results indicate that insider threats are significantly more
prevalent in distributed systems (34%) than in centralized ones (19%), with
critical CVSS scores (7.8–9.0). Man-in-the-middle attacks between nodes remain
dominant, with 68% caused by ineffective mutual authentication. AES-256-GCM
offers the best performance–security trade-off for data at rest, ML-KEM is
suitable for quantum-resistant use cases, while homomorphic encryption remains
impractical for production. Benchmarking shows a combined security overhead of
28–35% while maintaining regulatory compliance. The proposed framework provides
architecture-specific, compliant, and performance-aware security
recommendations, including applicability to Ukraine’s financial sector
regulations. |
|
Keywords: |
Cryptographic Methods, Authentication Mechanisms, Cyber Threat Taxonomy,
Consensus Algorithms, Security Architectural Framework |
|
DOI: |
|
|
Source: |
Journal of Theoretical and Applied Information Technology
30th April 2026 -- Vol. 104. No. 8-- 2026 |
|
Full
Text |
|
|
Title: |
ARTIFICIAL INTELLIGENCE IN THE FILM INDUSTRY: A STUDY ON THE PERSPECTIVE OF GEN
Z TOWARD AI-GENERATED CREATIVITY |
|
Author: |
AIKO MACHFUUDZOH, JENNIFER BRIGITTA TRISULO , SALSABILA ARUNDITA PRAMONO ,
YULIUS LIE |
|
Abstract: |
The film industry has changed with the rise of Artificial Intelligence in the
industry, especially in creative things like writing scripts, making special
effects and using fake actors. This study explores at what Generation Z, or also
known as Gen Z, perceives and accepts about Artificial Intelligence (AI) being
used to make things in Indonesian movies. From an Information Systems
perspective, however, technology acceptance research has predominantly focused
on productivity and organizational settings, leaving AI adoption in hedonic,
creativity-driven contexts underexplored, which a gap especially evident in
Indonesia, where the film industry is growing rapidly yet remains underexplore
in current IS research studies. This study extends the Technology Acceptance
Model (TAM) to examine how Generation Z perceives and accepts AI-generated
creativity in the Indonesian film industry. Using a quantitative approach with
PLS-SEM analysis using SmartPLS4 on 376 Gen Z respondents, to explore how
external influences, such as AI technology development and the business
environment, and internal influences, such as prior technology experience, in
shaping the perceptions of AI usefulness and task simplification, and how these
consequently affect attitudes, behavioral intention, and actual use. The
findings largely supported the study’s hypotheses, with Gen Z showing favorable
attitudes toward AI in filmmaking, particularly when it makes the creative
process more useful and manageable. That said, many respondents were
uncomfortable with AI fully replacing human creativity. The bottom line is
clear: AI is welcomed as a helper, not a substitute, and filmmakers who want to
bring AI into their work would do well to keep the human heart of storytelling
intact. |
|
Keywords: |
Artificial Intelligence, Film Industry, Technology Adoption, External & Internal
Factors, Perceived Usefulness, Task Simplification, Attitude Toward AI,
Behavioral Intention, Actual Use. |
|
DOI: |
|
|
Source: |
Journal of Theoretical and Applied Information Technology
30th April 2026 -- Vol. 104. No. 8-- 2026 |
|
Full
Text |
|
|
Title: |
CROSS-LAYER ATTENTION ADAPTATION FOR REAL-TIME NEURAL INFERENCE IN
EMBEDDED DEVICES |
|
Author: |
M. NAGABHUSHANA RAO , RAMESH BABU PITTALA , DAYANANDA R B , SANDHYA N , RAVI
KUMAR. M , Dr. PUSHPA R , DR. GRK PRASAD |
|
Abstract: |
There is limited processing capabilities, memory, and energy in edge devices
that work in real-time setting, which presents a major concern relating to the
deployment of deep neural networks on embedded systems. The growing demand of
low-latency, energy-efficient inference requires architecture revisions that
would not deplete computational resources to maintain accuracy. The proposed
paper presents a new framework Cross-Layer Attention Adaptation (CLAA) capable
of selectively activating or jumping neural layers in the inference process
depending on the complexity of input with the aim to mitigate computation
redundancy while sustaining performance. The suggested model utilizes the use of
lightweight attention controllers and dynamic gating processes to carry out
content-aware layer skipping on a modular network composition of convolutions.
It can be scaled on hardware constrained platform like Raspberry Pi 4B and STM32
microcontroller. Some of the primary findings on running benchmark experiments
on CIFAR-10 and the Tiny ImageNet would reveal that CLAA minimizes inference
latency by 45 percent and energy consumption by more than 36 percent with a
similar accuracy as typical full-depth CNNs. The visualisation plot such as pie
charts and confusion matrices as well as heat maps confirm the performance and
the interpretability of the model. The formulation of the edge AI by the CLAA
platform contributes a sustainable and deployable solution to the fledgling
landscape of adaptive neural architecture developers as well. It is of great
promise in intelligent sensing, mobile vision and timing-constrained automation,
especially in applications with tight computation and power requirements. |
|
Keywords: |
Cross-Layer Attention, Embedded AI, Layer Skipping, Real-Time Inference,
Energy-Efficient Deep Learning, Edge Computing. |
|
DOI: |
|
|
Source: |
Journal of Theoretical and Applied Information Technology
30th April 2026 -- Vol. 104. No. 8-- 2026 |
|
Full
Text |
|
|
Title: |
SOCSOH-HDL: HYBRID DEEP LEARNING FRAMEWORK FOR CONCURRENT ESTIMATION OF
STATE-OF-CHARGE AND STATE-OF-HEALTH IN LITHIUM-ION EV BATTERIES |
|
Author: |
ASHOK KUMAR BANDLA , Dr. GOPINATH PALAI , PROF. (DR.) RABI N SATPATHY |
|
Abstract: |
The SoCSoH-HDL framework is a hybrid Deep Learning model that can be used to
simultaneously estimate state-of-charge (SoC) and the state-of-health (SoH) of a
lithium-ion electric vehicle (EV) battery. The framework incorporates
one-dimensional convolutional neural networks (1D-CNNs) into Bi-LSTM/Transformer
components to target the short-term dynamics and recognises long-term
degradation trends by leveraging the multi-modal sensor measurements, the
physics-based features and the dual-headed hybrid temporal encoder. Use of
physics-informed diagnostics, including open-circuit voltage (OCV) mapping,
incremental capacity analysis (ICA), and direct current internal resistance
(DCIR) result in more interpretable diagnostics and limit predictions to real
world realistic values. More than 50,000 data points collected over a range of
temperatures, drive cycles, battery chemistries and noise conditions proved to
be very accurate with an average SoC RMSE value of 1.85% and SoH error value of
3.05 and a high R 2 value of 0.986. The structure can support real-time Battery
Management System (BMS) integration since it upholds sub-8 ms inference latency.
These findings indicate good generalization and high level of performance at
extreme operating condition, late aging of battery and average sensor noise. The
proposed solution is more superior to the existing situation of battery state
estimation processes because of the combination of short-term and long-term
estimation processes, improved accuracy evaluations, lowering the density of
computations, and enabling the deployment of EVs. |
|
Keywords: |
State-of-Charge Estimation, State-of-Health Prediction, Hybrid Deep Learning,
Lithium-Ion Battery Monitoring, Electric Vehicle BMS |
|
DOI: |
|
|
Source: |
Journal of Theoretical and Applied Information Technology
30th April 2026 -- Vol. 104. No. 8-- 2026 |
|
Full
Text |
|
|
Title: |
INTEGRATED CRIMINOLOGICAL-PSYCHOLOGICAL MODEL FOR PREDICTING CHILD CYBERBULLYING
RISKS IN DIGITAL ENVIRONMENTS |
|
Author: |
YULIIA BOIKO-BUZYL , VIKTORIIA HALCHENKO , IRYNA LUBENETS , VIRA KORKH , SERHII
PETROV |
|
Abstract: |
The research investigates a substantial investigation which unites
criminological and psychological approaches to develop an analytical system for
cyberbullying identification and prediction in digital environments. The
research established and tested a complete child-oriented cyberbullying risk
assessment model (CCRM) which evaluated how legal elements and psychological
elements and technical aspects work together to protect children from online
attacks. The research design unites Delphi-AHP expert weighting with correlation
and regression analysis and k-mean clustering and bootstrapping simulations (n =
500, α = 0.05) and Python 3.12 and Power BI and Tableau for computational
visualization. The research investigates Ukraine and Italy and Indonesia as test
sites because these countries have distinct legal frameworks and varying digital
infrastructure development. The Cyberbullying Risk Prediction Index (CRPI)
integrates five elements which include Legal and Institutional Protection (LIP)
and Psychological Risk Factors Index (PRF) and Technology Readiness (TDR) and
Criminological Response (CR) and Digital Literacy and Awareness (DLA). The
research findings indicated that Italy achieved the highest CRPI score at 0.78
while Ukraine developed a hybrid system with 0.66 and Indonesia used a
regulation-based model with 0.59. The research findings demonstrate that
psychological risk factors (PRF) and technical preparedness (TDR) determine
cyberbullying prevalence at 0.67 (R² = 0.67) while digital literacy (DLA) shows
a negative relationship with psychological vulnerability (PRF) at r = -0.72 (p <
0.01). The system provides practical value because it enables countries to
evaluate their digital security plans and identify cyber threats early and
maintain their policies in line with UN Convention on the Rights of the Child
and Budapest Convention on Cybercrime and EU General Data Protection Regulation
(GDPR 2016/679). |
|
Keywords: |
Cyberbullying, Digital Environment, Criminological Security, Psychological
Support, Digital Literacy, Rule of Law, Mental Health, Child Protection |
|
DOI: |
|
|
Source: |
Journal of Theoretical and Applied Information Technology
30th April 2026 -- Vol. 104. No. 8-- 2026 |
|
Full
Text |
|
|
Title: |
WHAT AFFECTS THE ACCEPTANCE OF ARTIFICIAL INTELLIGENCE (AI)-POWERED CREATIVITY
SUPPORT TOOLS AMONG THE INDUSTRIAL DESIGN COMMUNITY IN CHINA—A QUALITATIVE STUDY |
|
Author: |
JINZHI ZOU , KHAIRUL MANAMI KAMARUDIN, SAZRINEE ZAINAL ABIDIN , JING LIU |
|
Abstract: |
Artificial intelligence-based creative support tools (AI-CSTs) have currently
been progressively adopted in the creative industry. These tools demonstrate
significant potential for enhancing creative efficiency and optimizing workflow
processes. Yet, the unique acceptance dynamics of AI-CSTs in the industrial
design field, as opposed to tools in graphic design, constitute a significant
and unexamined research gap. This paper aims to explore the key factors
influencing the acceptance of AI-CSTs among the industrial design community.
This study employed in-depth interviews, conducted via online platform and
face-to-face communication, with 18 industrial design stakeholders possessing
over three years of professional experience. All interview content was
transcribed, coded, and analyzed using NVivo 12.0, with a thematic analysis
approach adopted to identify the primary influencing factors. The findings
reveal that users’ acceptance of AI-CSTs is influenced by multiple factors,
including the perception level (performance expectancy, effort expectancy,
perceived risk), individual level (technology optimism, innovativeness), and
technical level (facilitating conditions, price value, interactivity).
Understanding users’ attitudes and concerns regarding the use of AI-CSTs is
critical for both technology developers and policymakers to optimize tool design
and promotion strategies. This study contributed to enhancing the acceptance and
practical effectiveness of AI-CSTs. |
|
Keywords: |
Artificial Intelligence; Creativity Support Tools; Technology Acceptance;
Industrial Design; Qualitative Research |
|
DOI: |
|
|
Source: |
Journal of Theoretical and Applied Information Technology
30th April 2026 -- Vol. 104. No. 8-- 2026 |
|
Full
Text |
|
|
Title: |
DATA-DRIVEN OPTIMIZATION MODEL FOR FINANCIAL MARKET DYNAMICS ANALYSIS WITH
SOCIAL INFORMATION RETRIEVAL PERSPECTIVE |
|
Author: |
AZIZ BOUJEDDAINE, HAMID KHALIFI , YOUSSEF GHANOU , ACHRAF CHAKIR BARAKA ,
MOHAMED ATIFI , SARA RIAHI |
|
Abstract: |
Recent advances in financial research highlight the growing importance of
data-driven approaches, particularly Social Information Retrieval (SIR) and
Machine Learning perspectives, in understanding financial market behavior during
periods of uncertainty. These approaches emphasize the role of information
flows, investor perception, and market reactions derived from large-scale
financial and digital data.
The production and consumption of building
materials is a fundamental indicator in assessing the economic health of a
country and can also be considered a sensitive indicator of macroeconomic
fluctuations and crisis transmission in financial markets. In this context, we
analyze the financial impacts of the COVID-19 pandemic on the Building and
Construction Materials sector by focusing on a portfolio composed of four
leading securities in Morocco. This selection enables a focused analysis of
market behavior under extreme economic conditions.
A preliminary analysis
revealed a decline in closing prices and traded volumes during the first and
second lockdown periods of 2020, reflecting both fundamental economic shocks and
shifts in market sentiment. From a methodological perspective, a data-driven
optimization approach is adopted to quantify risk-return trade-offs under
different crisis scenarios. Subsequently, Modern Portfolio Theory (mean-variance
optimization) is used to estimate the portfolio’s expected return in 2008 and
2020 over the same period, allowing for a comparison of optimal return and
volatility between the two crises. This comparison provides insights into the
structural differences between a financial crisis and a pandemic-induced shock.
The analysis is discussed from a Social Information Retrieval perspective,
offering an informational interpretation of market reactions without modifying
the underlying quantitative framework. All computations were performed using R. |
|
Keywords: |
Data-driven, Optimization, Social Information Retrieval; Markowitz Theory;
Financial Markets; COVID-19; Economic Crisis |
|
DOI: |
|
|
Source: |
Journal of Theoretical and Applied Information Technology
30th April 2026 -- Vol. 104. No. 8-- 2026 |
|
Full
Text |
|
|
Title: |
A COMPUTER VISION-BASED THYROID DISEASE PREDICTION USING AN EBOLA
OPTIMIZATION-ENHANCED DEEP LEARNING MODEL |
|
Author: |
DR. CH. VIJAYA KUMAR , SURESHBABU AMARTHALURI , SIRISHA.J, RVVN BHEEMA RAO,
CH.SABITHA, V. PURUSHOTHAMA RAJU, N.PRASANNA LAKSHMI, B SREENIVAS REDDY |
|
Abstract: |
The thyroid is an important gland that regulates metabolism through its
hormones. Two major thyroid diseases, hyperthyroidism and hypothyroidism, are
caused by abnormal thyroid function. Computer vision is becoming increasingly
important in the medical diagnosis of various diseases, providing automated
analytical models for accurate and rapid diagnosis of lesions. This paper
proposed to a computer vision-based thyroid disease prediction framework using
an Ebola optimization-based deep learning structure (EODLM). The three basic
phases of the suggested structure are pre-processing, feature extraction, and
classification. Use open-source databases and a normalisation strategy to
substitute missing values during the pre-processing stage. A three-step feature
selection approach, involving bidirectional feature removal, backward feature
elimination, and forward feature elimination is used during the feature
extraction stage. During the classification phase, present the EODLM, an
innovative deep learning model that integrates the Elman Recurrent Neural
Network (ERNN) with the Ebola Optimisation Algorithm (EOA). The ERNN's weight
parameters are optimised using the EOA, the suggested strategy was put into
practise in MATLAB, and the following statistical metrics were used to assess
its effectiveness: F-measure, sensitivity, kappa, accuracy, recall, specificity,
and precision. the suggested approach was contrasted with popular machine
learning methods, such as Support Vector Machine (SVM), Efficient Feature
Extraction Recurrent Neural Networks (EFERNN), Deep Belief Neural Networks
(DBN), and Artificial Neural Networks (ANN). The suggested EODLM performed
better than any other method, yielding the following outcomes: F-measure (0.95),
specificity (0.99), recall (0.93), accuracy (0.99), kappa (0.94), sensitivity
(0.96), and precision (0.94). A promising computer vision-based framework for
the prediction of thyroid disease is the suggested EODLM. It works better than
other well-known machine learning algorithms and attains excellent accuracy. |
|
Keywords: |
Thyroid Prediction, Deep Learning Model, Elman Recurrent Neural Network, Ebola
Optimization Algorithm, Bidirectional Feature Elimination, Forward Feature
Selection, And Backward Feature Elimination. |
|
DOI: |
|
|
Source: |
Journal of Theoretical and Applied Information Technology
30th April 2026 -- Vol. 104. No. 8-- 2026 |
|
Full
Text |
|
|
Title: |
TIGLEAFNET-Q: AN EXPLAINABLE HYBRID QUANTUM CLASSICAL FRAMEWORK FOR MULTI-CROP
LEAF DISEASE DETECTION WITH ENHANCED FEATURE REPRESENTATION |
|
Author: |
D. VETRITHANGAM, KARTHIKEYAN T , VIVEKANANDAN S J , ASHWINI BARBADEKAR,
VASUKIDEVI G, MUTHUKUMAR SUBRAMANIAN, ASWATHY S NAIR |
|
Abstract: |
This study addresses the critical need for accurate and interpretable plant
disease detection in sustainable agriculture. Existing deep learning models
often suffer from limited generalization across crops, insufficient modeling of
complex feature interactions, and a lack of explainability, restricting their
real-world applicability. To overcome these challenges, this work proposes
TIGLeafNet-Q, a hybrid quantum–classical framework for multi-crop leaf disease
detection. The proposed approach integrates an InceptionV3 transfer learning
backbone for deep feature extraction, followed by feature compression and angle
encoding into quantum states. A variational quantum circuit (VQC) is employed to
capture higher-order correlations among disease features, and Grad-CAM is used
to provide visual interpretability. The model is evaluated on large-scale tomato
(22,980 images) and cotton (2,204 images) datasets. Experimental results
demonstrate classification accuracies of 99.28% (tomato) and 97.23% (cotton),
outperforming existing methods. The proposed framework contributes a scalable,
explainable, and multi-crop capable solution, highlighting the potential of
hybrid quantum–classical learning in agricultural intelligence systems. |
|
Keywords: |
Transfer Learning; Tomato; Cotton; Leaf Disease Detection; Deep Learning; Hybrid
Quantum-Classical Learning; Angle Encoding; Variational Quantum Circuit |
|
DOI: |
|
|
Source: |
Journal of Theoretical and Applied Information Technology
30th April 2026 -- Vol. 104. No. 8-- 2026 |
|
Full
Text |
|
|
Title: |
THE EFFECT OF GOOGLE WORKSPACE UTILIZATION ON THE EFFECTIVENESS OF DIGITAL
COLLABORATION IN STUDENT GROUP PROJECTS |
|
Author: |
RAINER RAINER , YULIUS LIE , ENGGAL SRIWARDININGSIH |
|
Abstract: |
The rapid move toward online systems in higher education has accelerated the use
of cloud-based tools, with Google Workspace now a staple for student group
projects. However, the high-end features of these technologies don’t always
guarantee effective teamwork, largely due to persistent socio-behavioral
obstacles like social loafing. This research examined how the use of Google
Workspace influences the performance of digital collaboration between students
with a unique positioning of social loafing as a moderating factor. The research
design used was quantitative, where data were gathered using the online survey
(N=413 active university students in Indonesia who use Google Docs, Google
Sheets, and Google Slides regularly). The data analysis was performed by the
Partial Least Squares Structural Equation Modeling (PLS-SEM) through SmartPLS
4.0. Data findings show that the following variables make significant and
positive contributions to the effectiveness of digital collaboration: Perceived
Ease of Use (PEOU), Perceived Usefulness (PU), and Google Workspace Utilization
(GWU). More importantly, the moderation analysis demonstrates a certain
behavioral paradox: social loafing has a considerable negative impact on the
advantages of perceived usefulness, yet it does not moderate the actual use of
technology behavior. This suggests that passive students might often use the
platform, but they fail to contribute anything of substantive value, thus
worsening the team dynamics. The study adds to the body of knowledge in the
Technology Acceptance Model (TAM) by demonstrating that the process of adopting
technologies in collaborative settings should be accompanied by proactive
pedagogic interventions that would help prevent the issue of free-riding. |
|
Keywords: |
Google Workspace, Digital Collaboration, TAM, Social Loafing, PLS-SEM |
|
DOI: |
|
|
Source: |
Journal of Theoretical and Applied Information Technology
30th April 2026 -- Vol. 104. No. 8-- 2026 |
|
Full
Text |
|
|
Title: |
USE OF ARTIFICIAL INTELLIGENCE FOR COUNTERING DISINFORMATION AS A COMPONENT OF
STATE INFORMATION SECURITY |
|
Author: |
OLHA ANTIPOVA , VLADYSLAV VEKLYCH , RAISA PERELYHINA , NAZARIY ADAMCHUK ,
NATALIYA MARCHUK |
|
Abstract: |
The increasing scale and sophistication of disinformation campaigns have posed
significant challenges to state information security, particularly in the
context of digital transformation and hybrid threats. This study examined the
application of artificial intelligence (AI) as a legally compliant instrument
for detecting and countering disinformation within national and supranational
regulatory frameworks. The purpose of the research was to develop and validate
an integrated model combining AI-based detection with legal-taxonomic
classification and compliance evaluation. The study employed an
interdisciplinary methodology that included legal analysis, natural language
processing (NLP), and compliance modelling. A multilingual dataset consisting of
8,000 text units was analysed using a BERT-based model, while disinformation
categories were structured according to Ukrainian legislation, European Union
law, and Council of Europe standards. The system was further evaluated through a
compliance framework measuring the legal validity of AI-generated actions. The
results demonstrated high technical and legal performance. Classification
reliability reached 94.68%, while detection effectiveness achieved an F1-score
of 94.13% across Ukrainian and English texts. The efficiency of compliance
amounted to 91.27%, confirming the ability of the system to generate legally
valid responses, including content removal requests and sanction
recommendations. The findings indicated that AI systems can operate consistently
across different legal regimes without significant loss of accuracy. The study
concluded that the integration of AI into legally structured information
security systems significantly enhances the effectiveness of disinformation
detection and mitigation while maintaining compliance with regulatory standards.
The proposed model contributes to bridging the gap between technological
innovation and legal applicability. Future research should focus on improving
explainability of AI models, harmonizing cross-jurisdictional legal frameworks,
and developing hybrid decision-making systems combining automated analysis with
human oversight. |
|
Keywords: |
Disinformation, Artificial Intelligence, Information Security, Legal Compliance,
Natural Language Processing, Cybersecurity, Digital Services, Machine Learning |
|
DOI: |
|
|
Source: |
Journal of Theoretical and Applied Information Technology
30th April 2026 -- Vol. 104. No. 8-- 2026 |
|
Full
Text |
|
|
Title: |
A RISK-AWARE COGNITIVE DECISION-MAKING FRAMEWORK FOR ENHANCING SECURITY IN
MODERN E-COMMERCE SYSTEMS |
|
Author: |
DR.N.ROOPALATHA , DR. VEERA ANKALU. VUYYURU , DR.S.GOKILAMANI , M. MISBA4, DR.
ARADHANA SAHU5, KULDOSHEV ILYOS SHUKHRATOVICH , A. Z. KHAN |
|
Abstract: |
The rapid growth of e-commerce has significantly increased exposure to
sophisticated security threats, including account takeovers, payment fraud, and
behavioural anomalies. Existing security mechanisms, such as static rule-based
systems and traditional machine learning classifiers, often fail to balance
detection accuracy with user convenience, leading to high false-positive rates
and poor customer experience. Moreover, these methods rarely account for
temporal patterns in user behaviour or adapt dynamically to evolving attack
strategies. To address these limitations, this study proposes a Risk-Aware
Cognitive Decision-Making Framework (RACDF) that integrates temporal deep
learning, cognitive user modelling, and adaptive intervention policies. The
framework employs a Temporal Fusion Transformer (TFT) to capture multivariate
sequential dependencies in transactional, behavioural, and device/network data,
producing accurate real-time risk scores. A Bayesian cognitive model
personalizes risk thresholds by representing individual user trust and
uncertainty, while a reinforcement learning-based policy engine selects optimal
interventions—ranging from soft prompts to transaction blocking—balancing
security and usability. The methodology follows an end-to-end workflow: data
collection and pre-processing, temporal feature embedding, TFT-based risk
prediction, cognitive decision modelling, RL-driven intervention selection, and
continuous feedback-driven learning. Evaluation on historical e-commerce
transaction datasets and simulated attack scenarios demonstrates that RACDF
outperforms conventional ML and static-rule baselines, achieving higher true
positive detection rates, lower false positives, and reduced user friction. This
study contributes a novel integration of temporal deep learning with cognitive
and reinforcement learning-based decision-making for e-commerce security. The
proposed approach provides a scalable, adaptive, and user-aware solution that
improves threat mitigation while maintaining a seamless shopping experience. |
|
Keywords: |
Cognitive User Modelling, E-Commerce Security, Reinforcement Learning,
Risk-Aware Decision Making, Temporal Fusion Transformer |
|
DOI: |
|
|
Source: |
Journal of Theoretical and Applied Information Technology
30th April 2026 -- Vol. 104. No. 8-- 2026 |
|
Full
Text |
|
|
|
