|
|
|
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
June 2026 | Vol. 104
No.11 |
|
Title: |
PROXY RE-ENCRYPTION WITH, BENCHMARKING, AND PHASED HYBRID MIGRATION FOR
TELEMEDICINE ARCHITECTURES |
|
Author: |
P.TEJASWINI , CH.NAGARAJU |
|
Abstract: |
The blistering development of the Internet of Medical Things (IoMT) and
telemedicine platforms has radically changed the healthcare delivery, providing
the opportunity to conduct remote monitoring, diagnose in real-time, and manage
electronic health records. Nevertheless, these developments present the serious
weaknesses in data protection, especially when it comes to ciphertext integrity,
unauthorized access, and the potential threat posed by quantum computing. In
this paper, a single framework has been provided to consider the short-term and
long-term cryptographic issues in healthcare data sharing. As a follow-up of a
blockchain-mimicking Proxy Re-Encryption (PRE) protocol and an extensive
post-quantum cryptographic (PQC) benchmarking analysis, we present a single
security architecture of IoMT settings. In the scheme based on PRE, identity
hash binding is also introduced when creating keys to ensure that there can be
verifiable connections between the identity of the user and the public keys to
improve accountability in data sharing between the Data Owners and the Data
Users. The transactions of blockchain are used to create a pairing-function
ciphertext verification scheme that is used to effectively stop the manipulation
of encrypted data stored on the cloud server. Accumulators that are managed by
smart contracts make it easy to manage user identities as well as perform
queries efficiently. At the same time, despite the fact that traditional
encryption protocols like RSA and ECC are becoming obsolete when quantum
adversaries use the Shor algorithm, the framework compares four PQC algorithms
that have been standardized by NIST Kyber, Dilithium, Falcon, and SPHINCS+. The
performance benchmarking indicates that Falcon has better encryption efficiency
of 17.16 ms with optimized storage capacity of 2.05 MB hence it can be found to
be highly suitable in the telemedicine applications that require low latency
whereas Kyber has a balance of speed and low computational overhead of 35.98.
One-way statistical analysis based on ANOVA helps prove that performance
differences are statistically significant between PQC algorithms. The evaluation
of the healthcare institutional preparedness indicates that technical expertise
and infrastructure capacity is a significant predictor of the success of PQC
adoption compared to budget allocation, with high-preparedness institutions
registering a score of 6.97/10 on both dimensions. The combined scheme entails a
computational efficiency improvement of the currently existing methods and will
cut down on the time of encryption, re-encryption, decryption, and re-decryption
by around 23.8, 71.4, 48 and 15.3 percent respectively and yet will not
compromise on the IND-ID-CPA security with the DBDH-assumption. All these
findings support a gradual hybrid cryptography migration plan, which includes
the introduction of quantum-resistant algorithms into the current IoMT systems
without interruption of care. [1,2] |
|
Keywords: |
Re-Encryption, Post-Quantum Algorithm, Post-Quantum Algorithm, Identity
Hash Binding |
|
DOI: |
|
|
Source: |
Journal of Theoretical and Applied Information Technology
15th June 2026 -- Vol. 104. No. 11-- 2026 |
|
Full
Text |
|
|
Title: |
TOKENIZATION-BASED TRUNCATING AND PADDING METHOD TO SOLVE LONG OPCODE SEQUENCE
PROBLEM IN LSTM FOR IOT MALWARE DETECTION |
|
Author: |
FIRAS SHIHAB AHMED, NORWATI MUSTAPHA, NOR FAZLIDA MOHD SANI, RAIHANI MOHAMED |
|
Abstract: |
Detecting malicious software in Internet of Things (IoT) datasets and
environments remain a significant challenge for researchers striving to secure
IoT networks. As the massive interconnection of Internet devices causes, most
previous research has explored deep learning approaches, particularly the LSTM
model, in detecting malware within operation codes (Opcodes) for ARM-based IoT
applications, owing to its strong classification capabilities. Despite the
advantages of using opcode features for detecting malicious software, the length
of the opcode sequence poses a major challenge for deep learning methods such as
the LSTM algorithm. Long opcode sequences can lead to information loss and
increased computational burden, which may result in the vanishing gradient
problem in the LSTM algorithm. To address this issue. This paper proposes a
tokenization-based truncating and padding method to solve this problem. It
shortens the length of the opcode sequence while keeping the classification
performance the same. This approach extracts a subset of opcode sequences based
on uniqueness, significantly reducing the length and size of the sequence while
preserving the quality of the dataset. The method was evaluated on three IoT
datasets from the Linux system and one dataset from the Windows system. The
findings show that, for all datasets, the suggested approach performs better
than current approaches in terms of accuracy and time. |
|
Keywords: |
Long Short-Term Memory (LSTM), Truncating and Padding, Cross-Validation, Malware
detection, Opcode. |
|
DOI: |
|
|
Source: |
Journal of Theoretical and Applied Information Technology
15th June 2026 -- Vol. 104. No. 11-- 2026 |
|
Full
Text |
|
|
Title: |
COGNITIVE CAPITAL AND DIGITAL TRANSFORMATION: AN EMPIRICAL INVESTIGATION
IN MOROCCAN SMES |
|
Author: |
SOUMAYA AHAROUAY, MONSIF BEN MESSAOUD LAYTI, RAFIA FRIJ, AHMED AHROUAY |
|
Abstract: |
This paper on Moroccan MSMEs makes an in-depth examination of CC for its impact
and enabling of the digital transformation journeys. The rapidly changing
environment of technological innovation and disruption that many firms face
presents a complicated and dual challenge; while there are also the increased
pressures and demands that will be encountered with this digitalization,
organizations cannot escape the adaptation to forces which may collaterally
constrain them through pre-existing inherited structures, routines or ways in
which they have been performing their operations traditionally. Consequently,
the current analysis investigates whether a range of internal cognitive and
organizational mechanisms (e.g. knowledge sharing, collective learning,
organizational adaptability, and proactive orientations toward change) are
indeed statistically linked to higher levels of digital maturity (and hence also
sophistication), and consequently can be said to induce such improvement in
performance.
A sound mixed-method research design was deliberately staunchly
embraced to provide answers to the above research objectives. Quantitative data
were carefully collected from the diverse sample of 150 Moroccan SMEs from
different industries and sectors, ensuring a broad and representative dataset.
In concert with this, qualitative data were collected through semi-structured
in-depth interviews with managerial profiles who play a crucial role in and are
accountable for leading their organizations digital transformation. These
interviews yielded rich contextual and often subtle insights into the processes.
Quantitative analyses were conducted in the statistical software package SPSS
(version 28.0) to conduct robust statistical testing and confirmation of
hypothesized relationships using these data.
The empirical results tangibly
demonstrate the positive and statistically significant connection between
overall intelligence of firms quantified as cognitive capital and level of
digital maturity achieved by these firms. In particular, organizations are more
likely to implement and realize benefits from complex digital solutions such as
enterprise resource planning (ERP), customer relationship management (CRM),
integration of different business applications in the cloud or on-premise if
they display heavier, deeper learning-related organizational capabilities and
more intensified knowledge exchange activities. These results provide strong
evidence that the extent to which smaller firms in Morocco successfully
undertake their digital transformation is not solely dependent on investing in
those new technologies, but rather it is highly reliant on an organizational
environment conducive for continuous learning, coordination and ability to
change effectively.
This study is an important contribution to the literature
based on objective preference linkage between cognitive capital mechanisms and
digital maturity, especially establishing this link in the concentration of an
emerging-economy setting. It also emphasizes the enduring role of knowledge
management as a core organizational logic, both in enabling continual
transformation and ensuring firm viability and sustainability in dynamic but
often unstable environmental contexts. |
|
Keywords: |
Cognitive Capital, Digital Transformation, Knowledge Management, Organizational
Learning, Moroccan SMEs. |
|
DOI: |
|
|
Source: |
Journal of Theoretical and Applied Information Technology
15th June 2026 -- Vol. 104. No. 11-- 2026 |
|
Full
Text |
|
|
Title: |
QUANTUM ACCELERATED REAL TIME ECG SIGNAL ANALYSIS FOR EARLY DETECTION OF CARDIAC
ABNORMALITIES |
|
Author: |
ANIL KUMAR PALLIKONDA, VENKATARAMANA BATTULA, CHINTALAPUDI RAKESH, T.SUDHA
RANI4, V. SWAPNA, VENKATESWARA RAO NARAMALA, APPIREDDY CHENNAKESAVAREDDY, RAVURI
DANIEL |
|
Abstract: |
Early and accurate identification of heart conditions from electrocardiogram
(ECG) signals is particularly important for ongoing patient health monitoring,
yet classical deep learning frameworks have been inadequate at achieving high
accuracy and low latency under noisy, real-time conditions. This work focuses on
developing a quantum-accelerated cardiac ECG (electrocardiogram) analysis. The
study introduces a novel hybrid quantum–classical framework capable of
simultaneously performing ECG denoising, feature embedding, and arrhythmia
classification with reduced latency and improved robustness under noisy
real-time conditions. The one introduced in the article is the hybrid
quantum-classical architecture, including Quantum Variational ECG Embedding
(QVEE) for high-dimensional morphological representation, Quantum Enhanced
Denoising Module (QEDM) for noise suppression and signal distortion, and hybrid
quantum classification for arrhythmia recognition. Experiments were conducted on
the MIT-BIH Arrhythmia Database, and a corpus with noise augmentation showed
that the proposed framework was found to be 99.4% correct with an F1 score of
0.97 and reduced the inference latency by 23.7% compared to state-of-the-art
CNN-LSTM, Transformer-based models, and showed higher robustness under the
condition of low signal-to-noise ratios. The results show that quantum
embeddings tend to improve ECG feature separability and that quantum denoising
helps preserve clinically relevant waveform structure (i.e., structure
detection), particularly for rare arrhythmias. The proposed framework is a
promising approach for establishing real-time quantum-assisted monitoring of the
human heart, enabling more reliable early diagnosis in wearable and clinical
environments. |
|
Keywords: |
Quantum-Accelerated ECG Analysis; Hybrid Quantum–Classical Model; Variational
Quantum Circuits; Real-Time Arrhythmia Detection; Quantum Denoising; Biomedical
Signal Processing. |
|
DOI: |
|
|
Source: |
Journal of Theoretical and Applied Information Technology
15th June 2026 -- Vol. 104. No. 11-- 2026 |
|
Full
Text |
|
|
Title: |
A NOVEL HYBRID FIREFLY ALGORITHM AND LSTM BASED INTELLIGENT SYSTEM FOR IOT
SECURITY |
|
Author: |
S R V PRASAD REDDY, K NARAYANA RAO, DIVESH SINGH SAI, MYLAVARAPU KALYAN RAM,
HANUMANTHA RAO BATTU, A MOHAN, PULICHERLA SIVA PRASAD, VASAVI MANDADI |
|
Abstract: |
The extensive adoption of the Internet of Things (IoT) has brought forth
numerous potential opportunities and advantages across various facets of our
lives. Nonetheless, it is unfortunate that IoT is also associated with various
vulnerabilities and a heightened risk of attacks and anomalies. The main
objective of these attacks is to unlawfully obtain sensitive information from
the system, while simultaneously creating interruptions in system access for
legitimate users. This study presents an enhanced Long Short-Term Memory (LSTM)
architecture aimed at effectively identifying attacks within an IoT environment.
The hyper-parameters of LSTM are optimized using an innovative Memetic Self
Adaptive Firefly Algorithm (MAFA). This study presented a perturbation operator
and incorporated it into the proposed MAFA to mitigate the risk of local optimum
solutions in the conventional firefly method. The comparative assessment of the
suggested methodology against other competing deep learning approaches reveals
that the proposed method excels across various performance metrics, including F1
score, F2 score, Fbeta score, precision, recall, ROC-AUC score, and accuracy.
The MAFA-LSTM methodology demonstrates exceptional performance compared to all
other approaches examined, achieving an accuracy of 99.99%. It demonstrates
exceptional effectiveness in precisely identifying intrusions within an IoT
setting. |
|
Keywords: |
IoT, IDS, TON-IoT, MAFA, DL, LSTM |
|
DOI: |
|
|
Source: |
Journal of Theoretical and Applied Information Technology
15th June 2026 -- Vol. 104. No. 11-- 2026 |
|
Full
Text |
|
|
Title: |
A NOVEL PRETRAINED U-NET AND DENSENET - BASED APPROACH FOR LUNG TUMOR
SEGMENTATION AND MALIGNANCY CLASSIFICATION IN CT IMAGES |
|
Author: |
DR. G LALITHA KUMARI, NALLA AKHILA, DR. K. KOTESWARA RAO, RAKESH KANCHARLA, N.
DEVI SRI, K. SREE VIDYA, N. RAMYA, M. SANJANA |
|
Abstract: |
Lung cancer is among the causes of cancer deaths in the world, and thus there is
a necessity to ensure that the computer tomography (CT) images of tumor are
accurately and early detected. Nonetheless, manual classification and evaluation
of lung nodules are time-consuming and subject to inter-observer errors because
of the complicated architecture, irregularity and low contrast of the tumors. In
order to overcome such problems, this paper attempts to offer a new deep
learning-based architecture that combines tumor segmentation with malignancy
classification in a single pipeline. This system has an advanced preprocessing
system based on Gaussian and Sobel filters, CLAHE, and lung field cutting to
enhance the details and isolate important areas. A fine-tuned U-Net model, using
ResNet-34 as the encoder, is used to produce high quality tumor segmentation
masks, which are then further improved with morphological operations and active
contour models. The divided tumor tissues are then dichotomized as benign or
malignant with the help of a fine-tuned DenseNet-121 network. Dice Similarity
Coefficient, Jaccard Index, sensitivity, specificity, Hausdorff distance and
computational time are used as performance measures. The results of the
experiment validate that the offered pipeline provides proper segmentation,
minimizes false positive, and produces credible malignancy prediction, which
makes the suggested pipeline a complete computer-aided tool to guide
radiologists in diagnosing early lung cancer and helping to make clinical
decisions |
|
Keywords: |
Lung Cancer, CT Images, Tumor Segmentation, UNet, Resnet-34, Classification,
Densenet-121 |
|
DOI: |
|
|
Source: |
Journal of Theoretical and Applied Information Technology
15th June 2026 -- Vol. 104. No. 11-- 2026 |
|
Full
Text |
|
|
Title: |
ELHE: AN EFFICIENT AND RELIABLE LIGHT WEIGHT HOMOMORPHIC CRYPTOGRAPHIC ALGORITHM
FOR EDGE COMPUTING |
|
Author: |
Dr M.V.R JYOTHISREE, K. SATHISH, Dr D.BHAVANA, Dr. P.SYAMALA RAO, Dr. HARI
JYOTHULA, Dr. SUBBA RAO POLAMURI, MANGALAGIRI SRIKANTH KUMAR, Dr. P.ANANTHA
LAKSHMI |
|
Abstract: |
This paper presents ELHE (Edge-optimized Lightweight Homomorphic Encryption), a
novel lightweight homomorphic cryptographic algorithm designed to solve the
problem of secure computation on resource-constrained edge devices — enabling
encrypted data to be processed without decryption, preserving confidentiality
end-to-end even on microcontroller-class hardware. The basis of ELHE is a
simplified instance of the Ring Learning With Errors (RLWE) problem, a widely
used hardness assumption in lattice-based cryptography. By reducing RLWE
parameters and streamlining noise control, ELHE minimizes computational overhead
while maintaining robust security guarantees. The algorithm has a clever
parameter tuning approach that balances security levels against performance
requirements carefully, such that even in lower configurations, a 128-bit
security level is guaranteed. This architecture makes ELHE amenable to being
deployed on average edge devices like the Raspberry Pi 4 and ESP32, where
conventional homomorphic encryption approaches would be impractical because they
are computationally and memory intensive.
Performance testing emphasizes the
performance of ELHE, demonstrating that it attains a 62% decrease in
computational overhead than conventional homomorphic encryption schemes.
Homomorphic addition and multiplication operations took 3.2 milliseconds and
18.7 milliseconds, respectively, on actual edge hardware. ELHE also uses merely
7.4% of available memory resources, thus rendering it very suitable for limited
computing environments. These advances show that high-assurance data privacy
using homomorphic encryption is possible even in low-resource systems, without
the need for high-power centralized infrastructure.
With respect to security,
ELHE is still strong against well-known cryptanalytic techniques used for
lattice-based schemes. The smaller parameter sizes are well designed to defend
against algebraic, statistical, and side-channel attacks while yet facilitating
practical deployment. The design also accommodates flexible deployment to enable
developers to tailor encryption strength and performance to match particular
application requirements. ELHE therefore embodies a proactive strategy to ensure
edge computing, providing a trustworthy cryptographic solution that addresses
both the performance needs and security requirements of next-generation
computing systems. |
|
Keywords: |
Homomorphic Encryption, Edge Computing, Lightweight Cryptography, Ring Learning
with Errors, Resource Optimization, Privacy-Preserving Computation,
Lattice-Based Cryptography |
|
DOI: |
|
|
Source: |
Journal of Theoretical and Applied Information Technology
15th June 2026 -- Vol. 104. No. 11-- 2026 |
|
Full
Text |
|
|
Title: |
AI-ENABLED CERVICAL CANCER RISK PREDICTION USING YOLOV12 DEEP FEATURES AND
OPTIMIZED RF-XGBOOST CLASSIFICATION |
|
Author: |
Dr. S. FOUZIA SAYEEDUNNISA, Dr. MANIZA HIJAB, Dr.S.GOMATHI, MYLAVARAPU KALYAN
RAM, SHAIK MABASHA, Dr.GARLAPATI NARAYANA, Dr.VANKUDOTHU MALSORU, Dr. SIVA KUMAR
PATHURI |
|
Abstract: |
Cervical cancer is the second malignancy that is highly diagnosed among women
and is regarded as one of the preventable ones. However, despite the existence
of effective prevention measures, the literature has shown that the level of
ignorance is high concerning cervical cancer, its causative factors, and
prevention measures in the developing countries, in particular. The medical
students and health care professionals are in a strategic position to raise the
awareness of people by elevating the level of their knowledge regarding the
symptoms, risk factors, and the early screening practices. Approximately 500,000
new cases of cervical cancer are reported each year all over the world with over
300,000 deaths. The main etiological factor leading to the development of the
disease is the persistent infection with the human papillomavirus (HPV) of
high-risk subtypes. Although cervical cancer cases can be considerably avoided
due to the organized screening and HPV immunization campaigns, almost 90 percent
of the cases occur in the states with low and middle income due to the lack of
access to systematic screening and immunization. By contrast, the cervical
cancer incidence and mortality rates in high-income countries have decreased
tremendously during the last thirty years due to the complex screening and early
intervention efforts. Also, fertility saving surgical procedures are now the
norm in the treatment of women with cervical cancer at the early and low risk
stages. The current paper introduces an advanced ensemble-based deep learning
framework that could be utilized to identify and forecast at an early stage
cervical cancer risk using both YOLOv12 and RF+XGboost to extract and classify
deep features respectively. The suggested YOLOv12-RFXGBoost model has performed
better than the conventional classifiers, with the peak accuracy of 96.5, and
can be used in the early detection of risks and the provision of clinical
decisions. And when compared with some base classifiers like Decision Tree and
Support Vector Machine the proposed classifier has given the best accuracy in
predicting/detecting cervical cancer. |
|
Keywords: |
Cervical cancer, Risk factors, screening, (HPV), YOLOv12, Random Forest,
XGBoost, Decision Tree, Support Vector Machine |
|
DOI: |
|
|
Source: |
Journal of Theoretical and Applied Information Technology
15th June 2026 -- Vol. 104. No. 11-- 2026 |
|
Full
Text |
|
|
Title: |
GPU-ACCELERATED QUANTUM HYBRID MODEL FOR SCALABLE AND EFFICIENT MOVIE
RECOMMENDATION |
|
Author: |
S.V.S.S.LAKSHMI, G. LAVANYA DEVI |
|
Abstract: |
The exploding interest in digital streaming services has escalated the need to
have readily scalable and accurate movie recommendation systems that can cope
with sparse data along with attractions surrounding the cold-start problem. This
paper proposes a GPU-accelerated quantum-infused recommendation system, which
includes quantum-inspired feature representations and classical deep learning
inference to increase predictive performance and faster computation. The
framework takes advantage of GPU parallelism to overcome the scalability
bottlenecks and make training and inference on large-scale and real-world movie
datasets faster. The proposed system is critically tested on the sparse and
cold-start small scale and compared with the state-of-the-art deep learning
recommender systems. Of great significance was the fact that the accuracy of its
GPU-accelerated YOLO+RFXGB-based feature extraction module came to 97.2%, which
was higher than conventional architectures, including ResNet-50(92%) and RegNetY
(89%), and this provided evidence of the usefulness of hybrid quantum-classical
modelling. The results obtained through the experimental part verify that the
quantum-infused approach using the GPU presents the best performance regarding
the ability to correctly match a movie and the person, the robustness of the
approach, and the scalability of the approach as the next generation of the
solution to the personalized movie recommendation task. |
|
Keywords: |
Quantum-Computing, GPU, Movie-Recommendation, , Sparse Data Handling, YOLO-Based
Feature Extraction, ResNet-50, RegNetY. |
|
DOI: |
|
|
Source: |
Journal of Theoretical and Applied Information Technology
15th June 2026 -- Vol. 104. No. 11-- 2026 |
|
Full
Text |
|
|
Title: |
FLOWER POLLINATION OPTIMIZATION APPLICATION IN WSN CLUSTERING AND
COMPRESSION-AWARE TRANSMISSION FOR IOT-DRIVEN ENVIRONMENTAL MONITORING |
|
Author: |
KAVITA AGRAWAL, DR. SHISH AHMAD |
|
Abstract: |
Wireless Sensor Networks (WSNs) are the components that lie at the fundamental
part of Internet of Things (IoT). They are used in systems of environmental
monitoring applications. The deployment in practical situations is limited
within the constraints of limited battery life and excessive overhead of data
communication. This paper presented a research work that focus on addressing
these challenges by integrating a framework of optimization process that is
associated for consideration of energy-aware selection of mechanism of sets of
CH and data compression prior to data transmission towards base station (BS).
The Flower Pollination Algorithm (FPA) as optimization method is followed for
finding optimal set of CHs. FPA use multiple network parameters that include
characteristics like node distribution, residual energy, divergence,
convergence, centrality, and distance to the BS. The data compression is applied
after the aggregation at CHs by applying simple hence fast compression methods
to add on higher efficiency. Two data compression algorithms are applied known
as ASWDR and Zstandard (ZSTD). Under the compression part the data size is
reduced prior to the transmission to BS from CHs. The simulation model developed
on MATLAB software is executed to generate results that are demonstrate the
performance enhancement observed by proposed optimal CHs set selection by FPA
associated with ASWDR or ZSTD based compression of data that results in reducing
the consumption of energy significantly. In this way the proposed work in this
article is introducing the performance enhancement respective to fast
convergence, network stability, load balancing, and higher network lifetime
compared to conventional methods like LEACH, SEP, DEEC and recent optimization
methods like GA, PSO based routing protocols for efficient WSN-based IoT
applications.
Research Gap: Existing literature focus either
optimization-based cluster head selection or data compression independently in
WSN. The existing routing protocols emphasize energy-efficient clustering but
ignore the large communication overhead caused by transmission of aggregated
environmental data. Similarly, compression techniques are rarely integrated with
routing frameworks for weather-monitoring applications.
Problem Addressed:
This study is addressing problem of high energy consumption that reduces network
lifetime in IoT-based WSNs system. It basically occurs due to inefficient CH
selection and data transmission overhead. The proposed work integrates FPA for
optimal CH selection with ASWDR and ZSTD data compression techniques prior to
transmission towards BS.
Rational: IoT-based WSN has limited energy of node
and communication overhead is due to continuous environmental data transmission.
Existing methods mainly focus on either routing optimization or data compression
separately, which limits overall network efficiency and lifetime. |
|
Keywords: |
IoT, WSN, Data compression, Clustering, Optimization |
|
DOI: |
|
|
Source: |
Journal of Theoretical and Applied Information Technology
15th June 2026 -- Vol. 104. No. 11-- 2026 |
|
Full
Text |
|
|
Title: |
AUTOMATED DOMAIN ANALYSIS FOR SOFTWARE REUSE USING PACKAGE ABSTRACTIONS AND
GENERICS |
|
Author: |
Dr B. JALENDER, Dr. T. VENKATA RAMANA, BANOTH SAMYA, Dr L. KIRAN KUMAR REDDY,
Dr. KACHAPURAM BASAVARAJU, GUGULOTHU VENKANNA |
|
Abstract: |
Domain analysis aims to identify and design reusable components for families of
products, defining the necessary domain roles, processes, models, and
architectures. While existing literature provides guidelines and techniques for
domain analysis, our work takes a more practical approach by automating these
principles. We implemented a system designed to address the challenges of Design
for Reuse (DFR) in detail. We have developed reusability guidelines, which
include that a good package can be used to help facilitate a good reuse
experience, similar to what Python has developed for its usage. The concept of
package as a strong reuse mechanism can be accomplished using Ada methods and
designs through proper usage of private types, separating specification from
implementation, and using generics for the parameterization of the users
package. Each of these methods can be used to produce a viable reusability
product regardless of the language being used for the creation of the product.
In this instance of this methodology, all composite types have been categorized
into abstract data structures. |
|
Keywords: |
Software reuse, Domain Analysis, reusability, Data Structures, components, CBSE. |
|
DOI: |
|
|
Source: |
Journal of Theoretical and Applied Information Technology
15th June 2026 -- Vol. 104. No. 11-- 2026 |
|
Full
Text |
|
|
Title: |
GAE-GCN: A DEEP GRAPH LEARNING MODEL FOR POWER PREDICTION IN CMOS VLSI CIRCUITS |
|
Author: |
PERIYASAMY K, G. Y. Rajaa Vikhram |
|
Abstract: |
In very large-scale integration (VLSI) design, power consumption estimation has
become something very crucial, having a direct effect on reliability, energy
efficiency, management of thermal properties, and other performances at a higher
level. With such complementary metal-oxide-semiconductor (CMOS) technology
scaling and the complex nature of the integrated circuit, power prediction at an
early stage is now of prime importance for design optimization of circuits.
Everything with simulation was considered better in terms of power estimation,
but it has always been too slow for design iterations. In view of resolving
issues in current power estimation systems, this study introduces a novel power
estimation scheme using Graph Autoencoder (GAE) combined with Graph
Convolutional Network (GCN). Such a model exploits the graph nature of CMOS VLSI
circuits, where logic gates and their interconnections are treated as nodes and
edges, respectively. The GAE encodes the circuit graphs into low-dimensional
latent structural features with preservation of both local and global dependency
relationships, while the GCN learns these features to predict power consumption
at the circuit level with high accuracy. Initially, gate-level attributes like
gate types, flip-flops, inputs, and outputs were considered from the ISCAS’89
benchmark dataset for training and evaluating the model. Experimental results
signify the excellent performance of the proposed GAE-GCN model as it has
yielded a prediction accuracy with a regression coefficient of 0.9999, RMSE of
0.00010, and a correlation coefficient of 0.999. The results are compared with
existing models outlined in the survey for validation purposes. The results
comparison indicates that the developed GAE-GCN model outperformed all other
models. |
|
Keywords: |
VLSI Design Circuits, Power Prediction, GAE, GCN, ISCAS’89 Dataset |
|
DOI: |
|
|
Source: |
Journal of Theoretical and Applied Information Technology
15th June 2026 -- Vol. 104. No. 11-- 2026 |
|
Full
Text |
|
|
Title: |
PERSONAL DATA PROCESSING IN STATE INFORMATION SYSTEMS: ADMINISTRATIVE, LEGAL,
AND TECHNOLOGICAL REGULATION |
|
Author: |
OLEH PREDMESTNIKOV, VLADYSLAV VEKLYCH, IVANNA HORBACH-KUDRIA, IVAN SHUMEIKO,
VIKTORIIA KORETSKA |
|
Abstract: |
The administrative and legal regulation of personal data processing in state
information systems (SIS) is becoming increasingly important because of the
development of e-government and digital identification systems. The aim of the
study is to establish how legal instruments ensure lawful, transparent, and
secure data processing. The methodology includes comparative legal analysis,
documentary review, and case studies, with a focus on jurisdictions with
developed e-government.
Significant discrepancies between norms and practice
were identified, especially in the areas of accountability, data minimization,
and cross-border exchange. In Germany, the BundID system provides legal
certainty thanks to clear obligations enshrined in federal law. Legal
transformation is ongoing in Ukraine (the Diia platform). Most systems lack
effective oversight and are not adapted to technological changes. The user
access to the X-Road-based platform complies with Articles 5 and 6 of the
General Data Protection Regulation (GDPR) regarding the lawfulness of processing
and identification. The Delphi procedure revealed only 12% agreement on the
criteria of minimization and user autonomy.
It is necessary to update the
legal framework, implement risk-based control, unify standards with
international norms. Research into automated tools and adaptive management
models using artificial intelligence (AI) and biometrics is promising. |
|
Keywords: |
Administrative Regulation, Legal Framework, Personal Data, Data Processing,
Government Systems, Information Systems, Public Administration, Data Protection |
|
DOI: |
|
|
Source: |
Journal of Theoretical and Applied Information Technology
15th June 2026 -- Vol. 104. No. 11-- 2026 |
|
Full
Text |
|
|
Title: |
A COMPUTATIONAL FRAMEWORK FOR IMPLEMENTATION OF E-GOVERNANCE IN DEVELOPING
COUNTRIES: AN ANALYTICAL CASE STUDY OF NEPAL |
|
Author: |
SANT KUMAR VERMA, VAISHALI SINGH |
|
Abstract: |
This study addresses the persistent challenges in implementing e-governance in
developing countries, with a focus on Nepal where infrastructural limitations,
institutional fragmentation, and low digital literacy continue to hinder
effective adoption. The research aims to identify key determinants influencing
e-governance implementation and to propose a computational framework that
integrates technological, institutional, and human dimensions. Using a
quantitative survey of 250 respondents across Nepal’s seven provinces, the study
applies multiple regression analysis to examine relationships among critical
factors. The findings reveal that institutional support [β = .35, p < .001] and
digital literacy [β = .29, p < .001] are the most significant predictors,
explaining 63% of the variance in adoption [R² = .63]. Based on these insights,
a four-layer computational framework is proposed. The study concludes that
effective e-governance implementation requires coordinated institutional
reforms, enhanced digital capacity, and inclusive infrastructure development.
The findings contribute both theoretically and practically by offering a
scalable model for developing nations. |
|
Keywords: |
E-Governance, Computational Framework, Nepal, ICT, Developing Countries, Digital
Literacy, Institutional Support. |
|
DOI: |
|
|
Source: |
Journal of Theoretical and Applied Information Technology
15th June 2026 -- Vol. 104. No. 11-- 2026 |
|
Full
Text |
|
|
Title: |
MAGNITUDE AND PENALTY-BASED PRUNING FOR QUANTIZATION WITH ZERO-SHOT APPROACH |
|
Author: |
PRIYANGA K.K, S. SABEEN |
|
Abstract: |
Deep convolutional neural networks deliver high predictive performance in image
classification. However, they often require substantial computation and memory.
This limits their use in resource-constrained environments. Many existing
compression methods apply pruning and quantization separately. They also focus
mainly on supervised learning with fixed label spaces. As a result, they may
cause accuracy loss, high optimization cost, and limited adaptability to unseen
classes. To address these limitations, this study proposes a hybrid Network
Pruning and Quantization framework integrated with zero-shot learning for
compressed deep convolutional neural networks. The proposed NPQ framework
combines magnitude-based pruning, penalty-based sparsity regularization, and
quantization-aware optimization within a unified design. It also preserves
semantic embedding consistency to support recognition of unseen classes. The
framework is evaluated on CIFAR 10, AWA2, and CUB benchmark datasets. Its
performance is compared with recent pruning and quantization methods including
LQ NET, RESREP, AutoPruner, and SIGMA. The experimental results show that NPQ
achieves competitive accuracy while reducing model complexity, memory usage, and
inference latency. Ablation studies further confirm that each component
contributes to compression efficiency and zero shot generalization performance.
Overall, the proposed NPQ framework offers an effective approach for building
compact and efficient deep learning models for real time and
resource-constrained deployment. |
|
Keywords: |
Deep Convolutional Neural Networks, Pruning, Quantization, Zero-Shot Learning,
Model Compression, Optimization |
|
DOI: |
|
|
Source: |
Journal of Theoretical and Applied Information Technology
15th June 2026 -- Vol. 104. No. 11-- 2026 |
|
Full
Text |
|
|
Title: |
SCALABLE HYBRID CLUSTERING FRAMEWORK VIA PARALLEL PARTICLE SWARM OPTIMIZATION |
|
Author: |
RAJASEKHAR KASEEBHOTLA, K.RAGHAVA RAO, MALLIKARJUNA RAO |
|
Abstract: |
Thus, an increasing number of data and their increasing complexity require new
approaches for clustering large data sets in many fields. This paper aims to
present a Scalable Hybrid Clustering Framework referred to as SHC-PPSO in this
paper based on Particle Swarm Optimization. It advances greatly. It is evident
from the parallel processing and PSO that SHC-PPSO optimally converges and
excels median attribute methods. This is faster than when data has to be
analyzed sequentially, which reduces the likelihood of making mistakes. For
reducing the curse of “Dimensality curse” actually, SHC-PPSO truncated the
processing power. Some MOA hybrid clustering algorithm enhancements increase
pattern recognition clearer and faster over the mean and the similarity distance
methods. It was clearly observed that proposed SHC-PPSO algorithm provided
better results than SCPSO-F1, SCPSO-F2 and PSOGSA with 65 & 95 % accuracy on
HIGGS and CICIDS2017 dataset respectively. SHC PPSO of distortions also enhanced
the performance and runtime. On the HIGGS dataset the overall running time was
decreased from 15000 seconds for a single node to 500 for 32 nodes while the
speedup almost followed the linear growth up to 30 for 32 nodes. These data
demonstrate that the system operates effectively and can be generalized to other
circumstances. Other areas that have benefited from SHC-PPSO are rivalry
mapping, biology, market categorization, and expansive network glaring
violations detection. To achieve efficiency for large, complex OBO datasets,
parallelization is done on precision clustering. With the help of computer
clustering the approach to data analysis and its interpretation could be changed
radically. |
|
Keywords: |
SHC-PPSO, Data Clustering, Parallel Particle Swarm Optimization,
High-Dimensional Data, Computational Efficiency |
|
DOI: |
|
|
Source: |
Journal of Theoretical and Applied Information Technology
15th June 2026 -- Vol. 104. No. 11-- 2026 |
|
Full
Text |
|
|
Title: |
MANAGEMENT INFORMATION SYSTEMS AND CRISIS MANAGEMENT IN HIGHER EDUCATION:
EVIDENCE FROM PALESTINE TECHNICAL UNIVERSITY-KADOORIE |
|
Author: |
QADRI K. ALZAGHAL |
|
Abstract: |
In an era characterized by constant technological progress and inherent
uncertainty, the importance of Management Information Systems (MIS) in building
organizational resilience has become a critical theme. Although previous studies
have investigated the connection between digital transformation, information
systems, and resilience in the context of higher education, few empirical
studies have examined the role that management information systems (MIS) play in
the various phases of crisis management, specifically crisis preparedness,
crisis response, and crisis recovery, in higher education institutions facing
conflicts. This research aims to explore the role of Management Information
Systems (MIS) in supporting crisis management at Palestine Technical
University-Kadoorie, with specific emphasis given to their contribution to
crisis preparedness, crisis response, and crisis recovery. The rationale behind
the current study stems from the premise that Palestinian higher education
organizations function in a setting characterized by continuous uncertainty and
vulnerability and subject to institutional restrictions, where successful
information exchange and coordination become essential factors for managing
crises effectively. This research adopts a quantitative research approach with a
cross-sectional research design. Data were collected through an online
questionnaire created using Google Forms. This research found that the proposed
model was supported by the data, with the results showing that MIS has a
positive and statistically significant impact on crisis preparedness, crisis
response, and crisis recovery. The findings also indicate that crisis
preparedness has a positive and statistically significant impact on crisis
response and that crisis response has a positive and statistically significant
impact on crisis recovery. The uniqueness of the research is seen in the fact
that the relationship between MIS effectiveness and three stages of crisis
management results has been described, thus revealing its contribution at
different stages. The findings further indicate that MIS is not only an
important organizational tool but also strategic for building organizational
resilience. Additionally, the study contributes new knowledge, as it shows how
management information systems (MIS) may increase resilience through information
visibility, rapid communication, coordination, and learning after crises in the
case of Palestinian higher education institutions. This research adds value to
the body of knowledge in MIS-driven crisis management through empirical analysis
conducted at Palestine Technical University-Kadoorie. The paper ends with
suggestions that can help enhance resilience within organizations in the higher
education sector facing uncertainties. |
|
Keywords: |
Management Information Systems; Crisis Management; Crisis Preparedness; Crisis
Response; Crisis Recovery; Higher Education; Palestine; Institutional
Resilience. |
|
DOI: |
|
|
Source: |
Journal of Theoretical and Applied Information Technology
15th June 2026 -- Vol. 104. No. 11-- 2026 |
|
Full
Text |
|
|
Title: |
FEDHCPDP: A FEDERATED SELF-OPTIMIZED MULTI-SCALE RESIDUAL ATTENTION GRAPH NEURAL
NETWORK FOR HETEROGENEOUS CROSS-PROJECT DEFECT PREDICTION |
|
Author: |
RADHALAKSHMI RAJAGOPALAN, DR. V. RADHA |
|
Abstract: |
Heterogeneous Cross-Project Defect Prediction (H-CPDP) plays a crucial role in
software quality assurance by predicting defects in projects with limited
historical defect information using knowledge transferred from external source
projects. However, existing H-CPDP approaches often suffer from feature
inconsistency, distribution mismatch, class imbalance, negative transfer, and
privacy concerns. To address these limitations, this study proposes a novel
Federated Heterogeneous Cross-Project Defect Prediction (FedHCPDP) framework
integrating ranking-guided project selection, federated learning, graph neural
networks, residual attention learning, and hybrid bio-inspired optimization. The
source projects from dataset are evaluated relative to the target project using
Cosine Similarity, Entropy Difference, and Wasserstein Distance to identify the
Top-K most relevant projects and reduce negative transfer. Furthermore, the
study proposed Self-Optimized Multi-Scale Residual Attention Graph Neural
Network (SOMRAGNN) to capture both local and global structural dependencies
among software modules through multi-scale graph convolution and residual
attention propagation. To further improve adaptive learning capability, the
attention coefficients and learning parameters of SOMRAGNN are optimized using a
Hybrid Secretary Bird–Bobcat Optimization Algorithm (HSBOA). The locally trained
models are aggregated using the proposed Federated Ranking-Guided Weighted
Averaging (FedRWAvg) mechanism, where highly relevant projects contribute more
significantly to the global model. Extensive experiments conducted on multiple
heterogeneous datasets demonstrate that the proposed FedHCPDP framework
consistently outperforms baseline models demonstrate the effectiveness, and
generalization capability of the proposed framework for reliable heterogeneous
software defect prediction. |
|
Keywords: |
Heterogeneous Cross-Project Defect Prediction, Federated Learning, Deep
Learning, Federated Heterogeneous Cross-Project Defect Prediction, Graph Neural
Network, Secretary Bobcat optimization Algorithm. |
|
DOI: |
|
|
Source: |
Journal of Theoretical and Applied Information Technology
15th June 2026 -- Vol. 104. No. 11-- 2026 |
|
Full
Text |
|
|
Title: |
SELFLIGHTNET: RESOLVING LOW-LIGHT SURVEILLANCE ENHANCEMENT WITHOUT PAIRED
TRAINING DATA |
|
Author: |
DR. S. SWAPNA RANI, DR. K. SHAILAJA, DUBBAKA SHIRISHA, VENKATESWARARAO PULIPATI,
VALIKI VIJAYABHASKER |
|
Abstract: |
Low-light image enhancement remains a critical challenge for intelligent
surveillance systems, where poor illumination severely impacts object
visibility, structural integrity, and downstream recognition tasks. Traditional
enhancement techniques, such as Histogram Equalisation and CLAHE, often
introduce noise amplification and fail to generalise across varying low-light
conditions. While offering significant improvements, recent deep learning-based
methods typically require paired low-light and normal-light datasets, which are
difficult to obtain in real-world surveillance environments. Furthermore,
existing self-supervised approaches often suffer from colour inconsistencies,
overenhancement, and low computational efficiency, hindering their applicability
to real-time deployment. To address these limitations, this paper proposes
SelfLight-Surv, a novel self-supervised framework that integrates the
SelfLightNet model to enable effective low-light enhancement without relying on
paired data. The architecture incorporates a Dual-Branch Illumination Prediction
(DBIP) module for learning local and global illumination patterns, a Dynamic
Exposure Correction Unit (DECU) for adaptive brightness adjustment, and a
Noise-Aware Fusion Layer to refine multi-scale feature representations. A
composite self-supervised loss function guides illumination consistency,
structural preservation, and contrastive feature learning. Experimental results
demonstrate that SelfLight-Surv outperforms existing methods, achieving up to
2.1 dB PSNR, higher SSIM, lower LPIPS scores, and significant gains in mean
Average Precision (mAP) for object detection tasks. The proposed framework also
achieves real-time inference speeds, making deployment in resource-constrained
surveillance systems practical. SelfLight-Surv offers a scalable, adaptable, and
efficient solution to enhance low-light images, significantly advancing
intelligent vision systems operating under adverse lighting conditions. This
work introduces a novel self-supervised low-light enhancement framework that
improves image quality and real-time surveillance performance without requiring
paired training data. |
|
Keywords: |
Low-Light Image Enhancement, Self-Supervised Learning, Smart Surveillance, Deep
Learning Framework, Image Quality Improvement |
|
DOI: |
|
|
Source: |
Journal of Theoretical and Applied Information Technology
15th June 2026 -- Vol. 104. No. 11-- 2026 |
|
Full
Text |
|
|
Title: |
A NOVEL HYBRID ADAPTIVE WEIGHTED EXTREME FUSION CLASSIFIER XFCM-JUSTICE FOR
COMPENSATION AWARDED IN MVOP LEGAL JUDGMENTS |
|
Author: |
MRS. PRASANNA KUMARI PATURU, DR. RAMESH BABU GAJJALA VENKATA |
|
Abstract: |
Motor Vehicle Accident Claim (MVOP) compensation prediction is a complex legal
analytics problem due to the heterogeneity of structured case attributes and
implicit judicial reasoning, leading to high variability and lack of consistency
in compensation outcomes. Existing machine learning models such as Random Forest
and Support Vector Machine, as well as several state-of-the-art approaches, fail
to generalize effectively across diverse legal scenarios and do not adequately
capture non-linear relationships in judicial data. To address this problem, this
study proposes XFCM-Justice, a novel hybrid machine learning framework that
integrates XGBoost, CatBoost, LightGBM, and ExtraTrees using a Hybrid Adaptive
Weighted Fusion (AWF) mechanism. AWF dynamically allocates the optimum weights
to every classifier depending on the distribution of their validation phase
error and macro F1 score, generating a context-sensitive and highly robust
fusion. The model is evaluated on a real-world MVOP dataset comprising 16
structured features, achieving superior performance with 94% accuracy along with
high precision, recall, F1-score, and ROC-AUC compared to base learners and
state-of-the-art models. The results demonstrate that adaptive fusion
significantly improves prediction stability, generalization, and class-wise
performance. The study concludes that XFCM-Justice is an effective and reliable
approach for legal decision-support systems in compensation prediction, offering
improved transparency and consistency in judicial analytics. |
|
Keywords: |
Motor Vehicle Accident Claims, Machine Learning, Compensation Prediction, Legal
Analytics, XGBoost, CatBoost, LightGBM, Ensemble Fusion, AWF Algorithm |
|
DOI: |
|
|
Source: |
Journal of Theoretical and Applied Information Technology
15th June 2026 -- Vol. 104. No. 11-- 2026 |
|
Full
Text |
|
|
Title: |
HYBRID ENSEMBLE-BASED DEEP LEARNING FRAMEWORK FOR VOLATILE CRYPTOCURRENCY PRICE
FORECASTING |
|
Author: |
D DAKSHAYANI HIMABINDU, ARCHANA KALIDINDI, V SRINADH, P JAHNAVI, B VASANTHA
RANI, B KRISHNA MOHAN |
|
Abstract: |
In today's world, cryptocurrency has emerged as a popular digital currency,
which is an alternative form of payment created using encryption algorithms.
Bitcoin is the most popular cryptocurrency. Bitcoin operates on a decentralized
computer network or distributed ledger that tracks transactions in the
cryptocurrency. Bitcoin is an innovative payment network and a new kind of
money. When computers on the network verify and process transactions, new
bitcoins are created, or mined. In return, for processing a transaction,
networked computers or miners receive a payment in Bitcoin. Demand for Bitcoin
has gone up over the past years. Bitcoin is very volatile and has multiple risks
like unpredictability, losing access to one's own money, security breaches,
being expensive and complex, and an uncertain future. Hence, developing a
machine learning model to accurately forecast future prices of Bitcoin using
machine learning algorithms is crucial. Many existing systems do not leverage
the strengths of different algorithms effectively, leading to suboptimal
predictive performance. This lack of diversity in model approaches often results
in poor generalization to unseen data and difficulty handling Bitcoin's high
volatility. The present study has developed an ensemble learning model by
integrating multiple algorithms, which include BiLSTM, XGBoost Regressor, and
Random Forest Regressor algorithms. The proposed ensemble model achieved an R²
score of 99.76%, demonstrating strong predictive performance. |
|
Keywords: |
Ensemble Learning, BiLSTM, XGBoost, Random Forest, Cryptocurrency Forecasting,
Volatility Analysis |
|
DOI: |
|
|
Source: |
Journal of Theoretical and Applied Information Technology
15th June 2026 -- Vol. 104. No. 11-- 2026 |
|
Full
Text |
|
|
Title: |
ENHANCING LUNG DISEASE DIAGNOSIS USING A CNN–CONVNEXT INTEGRATED DEEP LEARNING
FRAMEWORK |
|
Author: |
DR.K. SAI MADHURI, VISHAL BHARADWAJ MERUGA, V SITAMAHALAKSHMI, VENKAT RAO
PASUPULETI, YERRAVARAPU V V DURGA PRASAD, MR.ESWAR PATNALA, JAYAMMA RODDA, ARUN
KUMAR UNDAMATLA |
|
Abstract: |
Lung conditions detection at an early stage, like Tuberculosis, Pneumothorax,
COVID-19, and Lung Cancer, is vital for the fight against death. The manual way
of reading chest X-rays and CT scans takes a lot of time and has a high chance
of mistakes, which is why automated deep learning models are preferred. The
present study suggests a hybrid CNN–ConvNeXt model that merges powerful local
feature detection with global contextual learning. The public lung imaging
datasets were subjected to various preprocessing techniques like resizing,
normalization, augmentation, and lung-region segmentation. The training of the
models was done efficiently using AdamW optimizer, cosine learning rate
scheduling, and label smoothing techniques. ConvNeXt won the race with an
accuracy of 95–99% while CNN lagged behind, and it turned out to be a better
generalizer and more robust model. The system is capable of providing diagnostic
support that is scalable and suitable for clinical settings. Multimodal fusion,
severity estimation, edge deployment, and explainable AI for improved clinical
trust are some of the aspects that will be taken up in the future work |
|
Keywords: |
Lung Disease Classification, Deep Learning, CNN, ConvNeXt, Medical Imaging,
Chest X-ray, CT. |
|
DOI: |
|
|
Source: |
Journal of Theoretical and Applied Information Technology
15th June 2026 -- Vol. 104. No. 11-- 2026 |
|
Full
Text |
|
|
Title: |
ALGORITHMIC OPTIMIZATION OF MANAGEMENT PROCESSES IN THE PUBLIC SECTOR BASED ON
ARTIFICIAL INTELLIGENCE TECHNOLOGIES FOR INCREASING THE EFFICIENCY OF DIGITAL
GOVERNANCE |
|
Author: |
HENNADII MAZUR, OLEH MEDVYTSKYI, OLEKSII KHARYTONOV, ANATOLIY BABICHEV, OKSANA
MARUKHLENKO |
|
Abstract: |
The digitalization of public administration and the integration of artificial
intelligence (AI) technologies necessitate algorithmic optimization of
administrative processes and evidence-based assessment of their effectiveness in
digital governance systems.The aim of the study was to empirically verify the
model of algorithmic optimization of management processes in digital governance.
The research methodology combined process mining of administrative event logs,
digital twin modelling, quasi-experimental causal identification (DiD, ITS,
PSM), and statistical verification of results. This analytical scheme provided
an assessment of the effects of algorithmic optimization of management processes
in digital governance. Analysis of ≈50–150 thousand administrative events (24–36
months; 6–9 departments) revealed process asymmetry: the upper quartile of TAT
exceeded the median by 2.1–2.6 times, and 18–27% of cases generated more than
50% of delays. Quasi-experimental evaluation (≈300–500 panel observations)
recorded the effect of algorithmic optimization: TAT ↓9–14%, SLA-breach ↓11–18%,
rework ↓6–10%; robustness confirmed by bootstrap (B=1000–5000), FDR=0.05,
Durbin–Watson≈2, VIF<5. The study was the first to integrate process mining,
digital process twinning, and causal ML in a single empirical design analysing
≈50–150 thousand administrative events, providing quantitative verification of
the effects of algorithmic optimization in digital governance. Further research
should focus on longer panel designs (≥48–60 months) and experimental AI
interventions to test the scalability of algorithmic optimization and assess its
impact on the efficiency and quality of digital governance. |
|
Keywords: |
Algorithmic Optimization, Digital Governance, Artificial Intelligence, Public
Administration, Process Analytics, Digital Twin Of The Process,
Quasi-Experimental Analysis. |
|
DOI: |
|
|
Source: |
Journal of Theoretical and Applied Information Technology
15th June 2026 -- Vol. 104. No. 11-- 2026 |
|
Full
Text |
|
|
Title: |
CAN REINFORCEMENT LEARNING ENHANCE 3D AERIAL ROBOTIC EXPLORATION AND NAVIGATION
IN COMPLEX OBSTACLE-RICH ENVIRONMENTS? AN OPTIMISED SARSA-BASED LEARNING
APPROACH |
|
Author: |
VINOD K S , E D KANMANI RUBY |
|
Abstract: |
Aerial Robotic exploration and navigation play an important role in finding a
safe and optimum route from goal position to destination point. In this paper,
an Optimized State Action Reward State Action (SARSA) reinforcement learning
algorithm is proposed for autonomous navigation of UAVs in complex 3D obstacle
environments. A comparative study is carried out between existing traditional
SARSA algorithm and proposed optimized SARSA on real-time aerial robotic
application use traditional SARSA algorithm. The traditional algorithm exhibits
limitations in its performance in terms of convergence rate,
exploration-exploitation balance and computational complexity, whereas the
proposed algorithm includes parameters such as adaptive learning rate, decaying
epsilon-greedy policy and reward shaping which leads to improvement in learning
efficiency and navigation performance. To evaluate autonomous drone navigation,
a 3D grid-based simulation with spatial boundaries and dynamic obstacles has
been developed. The environment is designed to test flight performance in indoor
spaces, search and rescue situations and rough terrains. In this study, a
traditional SARSA agent with fixed hyper parameters is compared with a proposed
Optimised SARSA agent with adaptive learning mechanisms. The results shows that
the embedded optimisation technique greatly improve the reinforcement learning
performance in aerial robotics with 55% improvement in efficiency, 52 times
faster converge and 70% increase in stability. Future work will investigate
model-based approaches and deep reinforcement learning to address real-world
operational challenges. |
|
Keywords: |
Reinforcement Learning, SARSA, UAV Navigation, 3D Grid Simulation,
Hyperparameter Optimization |
|
DOI: |
|
|
Source: |
Journal of Theoretical and Applied Information Technology
15th June 2026 -- Vol. 104. No. 11-- 2026 |
|
Full
Text |
|
|
Title: |
DIGITAL TWIN TRAINING AND LEGAL GOVERNANCE FOR HUMANOID POLICE ROBOTS IN PATROL
AND ARREST-SUPPORT |
|
Author: |
SEUNGKOOK ROH |
|
Abstract: |
Humanoid police robots are emerging as Physical AI platforms because they can
operate in human-centered infrastructure such as stairs, doors, corridors,
vehicles, and control panels. This study is important because police robots do
not merely perform technical tasks; they may mediate state authority, personal
data processing, scene preservation, and possible restrictions on bodily
liberty. The goal of this paper is to design a digital twin-based training and
governance framework for humanoid police robots in patrol and arrest-support
operations in South Korea. The central hypothesis is that patrol functions can
be trained toward relatively high autonomy only if the digital twin encodes
legal permission states, privacy requirements, evidence preservation, and
human-command boundaries together with locomotion and perception; conversely,
arrest judgment, bodily restraint, and hazardous device activation should not be
automated under the current Korean legal framework. The study uses an
interdisciplinary design-science method combining a review of digital twin and
humanoid learning research, doctrinal analysis of Korean constitutional,
criminal procedure, police, privacy, AI, and compensation law, and synthesis of
a deployment-oriented architecture. The main findings are a five-layer digital
twin, an autonomy-permission matrix, a benchmark-and-phase-gate validation
matrix, and a human-in-command operating procedure. Practically, the framework
helps police agencies, vendors, and regulators convert broad safety and legality
requirements into testable design controls before public pilots. The paper
concludes that the most defensible near-term model is not an autonomous arrest
robot but a digitally trained police-support humanoid with law-in-the-loop
constraints, privacy-by-design controls, and audit-ready evidence logs. |
|
Keywords: |
Physical AI, Digital Twin, Humanoid Police Robot, Sim-to-Real Transfer, AI
Governance |
|
DOI: |
|
|
Source: |
Journal of Theoretical and Applied Information Technology
15th June 2026 -- Vol. 104. No. 11-- 2026 |
|
Full
Text |
|
|
Title: |
SECURING E-VOTING DATA USING PROPOSED LIGHTWEIGHT ARX BLOCK CIPHER ALGORITHM |
|
Author: |
SAFA A. AHMED , ALI M. SAGHEER |
|
Abstract: |
The Internet of Things applications have developed fast, leading to the need to
implement lightweight encryption schemes that can balance security with
resources efficiency in resource-constrained environments. This paper suggests a
lightweight ARX-based block cipher that combines dynamically generated
substitution boxes based on a six-dimensional chaotic system with an adjusted
key expansion algorithm based on AES. The plaintext is separated into two
parallel streams, and the work of each round consists of key addition, nonlinear
substitution by a dynamic S-Box, and ARX transformation in a symmetric
Feistel-like network. The suggested key scheduling mechanism minimizes inter-key
correlation and increases sensitivity to small key variations. The security
evaluation methods used are entropy, Hamming distance, correlation analysis,
NIST randomness tests, SAC and BIC tests. Experiments indicate that the proposed
framework is highly diffusive, random, and efficient in encryption and
decryption, thereby making it appropriate in protecting confidential database in
a constricted environment. |
|
Keywords: |
Lightweight Cryptography, ARX Block Cipher, Key Expansion, Chaotic system, S-Box |
|
DOI: |
|
|
Source: |
Journal of Theoretical and Applied Information Technology
15th June 2026 -- Vol. 104. No. 11-- 2026 |
|
Full
Text |
|
|
Title: |
DESIGN, DEVELOPMENT AND EVALUATION OF AN INTERACTIVE 2D ANIMATION-BASED LEARNING
SYSTEM FOR BAHASA MELAYU GRAMMAR ACQUISITION |
|
Author: |
SARNI SUHAILA RAHIM, MUHAMMAD FAKHRUL HADI FADZLI, SURIATI KHARTINI JALI |
|
Abstract: |
Animation has emerged as a significant asset for facilitating language
acquisition in the contemporary, rapidly digitalizing educational landscape, as
it enhances learner engagement and promotes comprehension and retention.
Conventional language teaching methods sometimes depend on fixed, text-based
materials, making abstract grammatical concepts difficult for students to
understand. Interactive 2D animation offers a more accessible and engaging
alternative by visually and meaningfully conveying the language curriculum,
allowing learners to connect theory with practical application. This research
examines the development of an interactive 2D animation instructional tool
designed to aid secondary school students in learning Bahasa Melayu, with a
particular focus on Golongan Kata. The animation’s effectiveness was assessed by
systematic testing with subject matter experts, students, and multimedia
professionals, focusing on learner engagement and language proficiency. The
results demonstrate that interactive animation significantly enhanced students’
interest, engagement, and academic performance. The study underscores the
potential of multimedia-based methodologies to enhance the engagement and
efficacy of Bahasa Melayu learning, advocating for a transition towards more
interactive and learner-centered methods in language education. |
|
Keywords: |
Interactive Animation, Language Learning, Bahasa Melayu Education, Digital
Learning Tools, Linguistic Proficiency |
|
DOI: |
|
|
Source: |
Journal of Theoretical and Applied Information Technology
15th June 2026 -- Vol. 104. No. 11-- 2026 |
|
Full
Text |
|
|
Title: |
SHAPE-PRIOR LEARNING FOR KIDNEY SEGMENTATION USING SHAPE-ORIENTED CONVOLUTIONAL
AUTO-ENCODER ENHANCED DEEP NETWORKS |
|
Author: |
K. BHAGYA REKHA, KAVILA MONI SUSHMA DEEP, SATHISH VUYYALA, K. DEVIPRIYA4, H K
PRASAD KATAMREDDI, BOYAPATI RAMADEVI6, KIRUTHIKA S, DR. JAMPANI SATISH BABU |
|
Abstract: |
The segmentation of kidneys is a challenging task in medical image analysis,
particularly for early diagnosis and treatment of renal disorders. Having a
clear, well-defined outline of kidney structures from computed tomography (CT)
and magnetic resonance imaging (MRI) helps doctors tremendously with diagnosis,
surgical planning, and beyond, as well as with monitoring disease progression.
But there are many problems, such as unevenly outlined boundaries, low image
contrast, mottled patterns, and normal anatomical variations that occur from one
patient to another, which all decrease the efficiency of the most common
segmentation techniques. Therefore, to circumvent these constraints, this study
proposes a framework, called Shape-Prior Learning, for kidney segmentation with
deep networks enhanced with SOCAE. In principle, the concept is to integrate a
convolutional auto-encoder based on shape, SOCAE, and a deep learning
architecture to retain anatomical consistency and allow the system to learn
structural cues and/or spatial characteristics of the kidneys. The pipeline
comprises image pre-processing, feature extraction, shape-prior learning, and,
at its end, the segmentation stage. The primary goal of the SOCAE part is to
obtain the local texture information and additional global kidney shape
representations. Therefore, the network can achieve higher segmentation accuracy
and fewer boundary errors, such as misclassifications. Including shape-prior
constraints in a deep network greatly reduces shape invariance in addressing the
challenging kidney region while preserving edges. Furthermore, features are
enhanced and normalized during training to ensure the model remains robust and
can be applied to different medical image databases. Finally, experimental
validation was performed on kidney benchmark image datasets using different
performance metrics, including Accuracy, Precision, Recall, Dice Similarity
Coefficient, and F1-Score. Comparing the SOCAE-enhanced deep network to the CNN
and the U-Net-based CNN, respectively, in an almost consistent manner, it can be
concluded that the proposed network achieved the highest accuracy. For accuracy,
the model has a high accuracy rate of 0.9882, precision of 0.9814, recall of
0.9776, and F1-Score of 0.9795. The segmentations are consistent. The Precision
improvement indicates fewer false-positive segmentation areas, and the overall
higher F1 Score suggests a fairly good balance between Precision and Recall.
Overall, these outcomes show the proposed Shape-Prior Learning approach
meaningfully boosts kidney segmentation, and it offers a practical method for
automated medical image processing, helping more intelligent clinical
decision-making systems as well as computer-aided diagnosis tasks. |
|
Keywords: |
Kidney Segmentation, Shape-Prior Learning, Shape-Oriented Convolutional
Autoencoder (SOCAE), Deep Learning, Medical Image Processing. |
|
DOI: |
|
|
Source: |
Journal of Theoretical and Applied Information Technology
15th June 2026 -- Vol. 104. No. 11-- 2026 |
|
Full
Text |
|
|
Title: |
INTEGRATING METAVERSE FOR THE RESTORATION OF NIAS CULTURE: AN APPROACH TO
HERITAGE PRESERVATION AND REGIONAL REVENUE OPTIMIZATION |
|
Author: |
ANNALISA SONARIA HASIBUAN, ANDAM LUKCYHASNITA, ORLI BINTA TUMANGGOR, ANDI
SUPRIADI CHAN |
|
Abstract: |
Globalization, modernization, and the limited digital documentation of
indigenous heritage have created an urgent need for information technology-based
approaches that can preserve, represent, and create sustainable value from local
culture. Existing metaverse-based heritage studies have mainly emphasized
virtual exhibitions or museum extensions, while fewer studies have integrated
architectural reconstruction, community validation, digital ownership, and
tourism-oriented value creation in one framework for a living indigenous
culture. This study addresses this gap by developing a participatory metaverse
framework for the restoration of Nias cultural heritage. The research design
combines human-centered design, ethnographic documentation, architectural
visualization, and applied computing. Data were collected through field
observations, interviews with local stakeholders, literature review, and
documentation of Nias traditional houses and village layouts. The development
process included basic 3D modeling, architectural detailing, material texturing,
environmental reconstruction, final rendering, and conceptual integration with
blockchain-based authenticity and asset ownership mechanisms. The results show
that Nias traditional houses and village compositions can be digitally
reconstructed as immersive and culturally meaningful environments. Compared with
prior museum-centered metaverse studies, the proposed framework contributes an
IT-based heritage restoration model that integrates 3D visualization,
participatory validation, blockchain-enabled authenticity, and digital tourism
potential. The practical implication is that local governments, cultural
institutions, and creative industries can use the framework to support heritage
education, virtual tourism, digital asset governance, and regional revenue
diversification. Limitations include the absence of large-scale user testing,
limited empirical measurement of economic impact, and the need for long-term
infrastructure validation. Future work should evaluate user experience,
community acceptance, and implementation readiness in real metaverse platforms. |
|
Keywords: |
Metaverse, Cultural Heritage, Nias, Digital Preservation, Virtual Tourism.
Commas |
|
DOI: |
|
|
Source: |
Journal of Theoretical and Applied Information Technology
15th June 2026 -- Vol. 104. No. 11-- 2026 |
|
Full
Text |
|
|
Title: |
CONSISTENT FRUITFLY OPTIMIZATION-BASED GRAPH NEURAL NETWORK (CFO-GNN) FOR
ANALYZING SENTIMENTS IN AUGMENTED REALITY-ENABLED ONLINE SHOPPING |
|
Author: |
PRAGATHI ARAVABOOMI, USHA S, 3NELSONMANDELA S, DURGESH TRIPATHI, BROSKHAN P |
|
Abstract: |
Online shopping has evolved significantly with the integration of augmented
reality (AR) technology, offering users the ability to visualize products in
their physical space before making a purchase. However, sentiment analysis
within AR-enabled platforms faces challenges due to sparse review data, unlike
traditional e-commerce platforms. The Consistent Fruitfly Optimization-Based
Graph Neural Network (CFO-GNN) proposed in this paper addresses this challenge
by combining fruitfly optimization with graph neural networks. This innovative
approach allows CFO-GNN to efficiently handle sparse data while capturing the
intricate relationships present in AR shopping experiences. By leveraging these
techniques, CFO-GNN enables more accurate sentiment analysis, empowering
businesses to make informed decisions and enhance user satisfaction in the
dynamic landscape of AR-enabled online shopping. Through comprehensive
evaluation on a diverse dataset, CFO-GNN demonstrates its effectiveness in
improving sentiment analysis within AR environments, highlighting its potential
to drive advancements in user experience and competitiveness for businesses
operating in AR-enabled online retail. |
|
Keywords: |
Augmented Reality, Analysis, Classification, Online Shopping, Sentiment, Sparse
Data |
|
DOI: |
|
|
Source: |
Journal of Theoretical and Applied Information Technology
15th June 2026 -- Vol. 104. No. 11-- 2026 |
|
Full
Text |
|
|
Title: |
SHAP-ENHANCED GENOMIC PREDICTION FOR TRANSPARENT AND TRUSTWORTHY PRECISION
MEDICINE |
|
Author: |
SHAIK MOHAMMAD RAFI, DR SARANGE SHREEPAD MAROTRAO, DR R YOGESH RAJ KUMAR,
GUNASUNDARI B, DR V. P. MURUGAN, AMIT VERMA, DR UVANESHWARI.M, DR R. SENTHAMIL
SELVAN |
|
Abstract: |
Precision medicine aims to apply genetic data specific to individual patients to
deliver an individual diagnosis and treatment plan. One of the typical obstacles
to the application of machine learning models in clinical practice is their
opaque nature, which is impressive in processing complex genetic data. To
overcome this limitation, this study proposes a genomic prediction framework
that is SHAP-enhanced. Such a framework would render the models more transparent
and understandable to use in therapy. In order to illuminate model decisions,
the predictive pipeline uses Shapley Additive Explanations (SHAP). It is a tool
that identifies and measures the effects of individual genetic differences on
individual predictions. The framework can be seen to be correct in a clinical
case study utilising patient-specific genomic data; SHAP can find significant
genetic biomarkers of the response to treatment and the risk of illness. The
findings of the experiments prove that the proposed approach preserves the high
degree of prediction accuracy and makes the outcomes easier to achieve and
understand, which is excellent news among physicians who need to understand how
the model managed to reach the results. Overall, our study demonstrates that
explainable AI methods, coupled with machine learning, can assist in reliable
biomarker discovery and increase the level of trust in Genomic-based clinical
decision-making. This, in its turn, may result in more patient-centred precision
medicine. In experimental settings, GenoGraphFormer consistently identified five
clinically actionable biomarkers with greater discrimination (AUROC = 0.913,
AUPRC = 0.764), improved calibration (Brier score = 0.121, ECE10 = 0.027), and
the best attribution stability (0.88) compared to all comparison models. The
results show that a strong, understandable framework for genomic clinical
decision support is produced by combining biological graph priors, transformer
attention, and consensus SHAP attribution. |
|
Keywords: |
Precision Medicine, Genomic Analysis, Explainable Artificial Intelligence (XAI),
SHAP, Biomarker Discovery, Clinical Case Study, Personalised Healthcare |
|
DOI: |
|
|
Source: |
Journal of Theoretical and Applied Information Technology
15th June 2026 -- Vol. 104. No. 11-- 2026 |
|
Full
Text |
|
|
Title: |
SEGMENTATION-ENHANCED DEEP LEARNING MODEL FOR ROBUST RETINAL DISEASE
IDENTIFICATION FROM OCT SCANS |
|
Author: |
SATHISH KAMALAKANNAN, Dr B. KIRUBAGARI, Dr J. JEGAN, Dr R. THIYAGARAJAN |
|
Abstract: |
OCT imaging is widely used for diagnosing retinal diseases. However, manual
evaluation is time-consuming and depends on expert ophthalmologists. This paper
proposes an automated deep learning approach that combines lesion localization
with disease classification. First, retinal lesion regions are segmented using
an improved U-Net architecture with multi-scale feature fusion, enabling precise
identification of affected portions of the retina. Next, the segmented lesion
maps are fed into a classifier network (ResNet/DenseNet) for retinal disease
classification. The fusion of segmentation-based lesion information improves
disease discrimination and reduces false predictions. The proposed method
demonstrates improved accuracy and efficiency over conventional end-to-end
classification models. The approach supports clinical interpretability by
clearly highlighting the affected retinal portion along with the disease label. |
|
Keywords: |
Retinal lesion segmentation, U-Net, OCT, Feature fusion, Deep learning,
Localization |
|
DOI: |
|
|
Source: |
Journal of Theoretical and Applied Information Technology
15th June 2026 -- Vol. 104. No. 11-- 2026 |
|
Full
Text |
|
|
Title: |
DISENTANGLE ATTENTION AND MASKED CUSTOM LARGE LANGUAGE MODEL BASED FAKE NEWS
DETECTION |
|
Author: |
K. SARITHA DEVI., DR. M. CHIDAMBARAM |
|
Abstract: |
False information presented as reliable news is known as fake news, and it's
dissemination undermines democracy and public confidence. Large Language Models
(LLMs) have two roles in this field: they can both propagate and identify false
information. Although BERT is a popular tool for detecting fake news, many other
approaches have trouble with intricate linguistic patterns and don't provide
clear justifications for their conclusions. We suggest a BERT-based model,
DAE-MCD-LLM, to enhance this. The method consists of three steps: news
classification, key term extraction, and data cleaning. The decoder adds more
layers to the enhance classification after the encoder captures word meaning and
position. When tested on the WELFAKE dataset, the model outperformed two
alternative methods in accuracy, precision, F1 score, and speed, demonstrating
its efficacy and reliability in detecting fake news. |
|
Keywords: |
Fake News Detection, Large Language Model, Disentangle Attention-based Encoder,
Masked Custom Decoder, WELFAKE dataset |
|
DOI: |
|
|
Source: |
Journal of Theoretical and Applied Information Technology
15th June 2026 -- Vol. 104. No. 11-- 2026 |
|
Full
Text |
|
|
Title: |
A FEDERATED LEARNING APPROACH FOR ROBUST HEART DISEASE RISK SCORING FROM
DISTRIBUTED DATA |
|
Author: |
Dr.T.KANIMOZHI, G. HEMANTH KUMAR YADAV, Dr. K. PRABHAVATHI, VENKATESWARLU
SUNKARI, SUDHEER BENARJI P, AMIRTHASARAVANAN ARIVUNAMBI, Dr.P.THIRUMOORTHY, Dr.
T. KARTHIKEYAN, Dr .JEEVAN JALA, Dr.BH.KRISHNA MOHAN, DR.T.VENGATESH |
|
Abstract: |
Heart disease remains the leading cause of mortality worldwide, with accurate
risk stratification being crucial for effective prevention and treatment.
However, the development of robust machine learning models for cardiovascular
risk prediction is severely hampered by the fragmentation of patient data across
multiple healthcare institutions and stringent privacy regulations (e.g., HIPAA,
GDPR) that prohibit centralized data sharing. To address this critical
challenge, this paper proposes FedCARDIA, a novel federated learning framework
with cross-institutional feature fusion for heart disease risk stratification.
Unlike conventional federated averaging methods that share model weights which
remain vulnerable to inversion attacks FedCARDIA shares only abstract
intermediate feature representations from local multi-modal data (longitudinal
electronic health records and echocardiogram images). A global classifier is
then trained on these aggregated features, enabling collaborative learning
without exposing raw patient data. Evaluated on a multi-institutional dataset
comprising 12,458 patients from three independent hospitals with non-IID data
distributions, FedCARDIA achieves an area under the ROC curve of 0.923 for
5-year cardiovascular risk prediction. This significantly outperforms
single-institution local models (AUC: 0.834–0.861) and standard Federated
Averaging (AUC: 0.882), while closely approaching the privacy-violating
centralized oracle model (AUC: 0.935). The framework maintains robust
performance across heterogeneous institutional data, offering a viable and
scalable pathway for privacy-preserving collaborative learning in healthcare
without compromising patient confidentiality. |
|
Keywords: |
Federated Learning, Heart Disease Risk Stratification, Privacy-Preserving
Machine Learning, Feature Fusion, Multi-Modal Learning, Cardiovascular
Informatics. |
|
DOI: |
|
|
Source: |
Journal of Theoretical and Applied Information Technology
15th June 2026 -- Vol. 104. No. 11-- 2026 |
|
Full
Text |
|
|
Title: |
CYBERBULLYING SEVERITY DETECTION IN MALAY-ENGLISH CODE-MIXED TEXT: A HYBRID
NLP-CNN APPROACH WITH USER BEHAVIOR PROFILING |
|
Author: |
IBRAHIM INUSA, RINA MD ANWAR, ASMIDAR ABU BAKAR, FIZA ABDUL RAHIM, MARINA MD
DIN, ALIZA ABDUL LATIF |
|
Abstract: |
Cyberbullying detection remains a significant and ongoing challenge,
particularly within multilingual code-mixed discourse such as Malay–English,
where existing approaches remain limited to binary identification and are
predominantly developed on monolingual datasets, overlooking both the varying
severity of cyberbullying and user behavioral patterns, which are essential for
developing effective interventions. Despite the growing prevalence of
cyberbullying, severity-aware detection for Malay–English code-mixed text
remains largely underexplored, leaving a methodological gap between real-world
multilingual communication patterns and current automated moderation systems.
Given the linguistic variability of code-mixed discourse and the need to capture
both lexical intensity and contextual semantic cues in Malay–English code-mixed
cyberbullying interactions, a hybrid modeling strategy that integrates
statistical lexical features and neural contextual representations provides a
more suitable framework for severity-aware cyberbullying detection. To address
this limitation, this study proposes a hybrid Natural Language
Processing–Convolutional Neural Network (NLP–CNN) framework designed to classify
cyberbullying instances across three severity levels (low, medium, and high)
while simultaneously profiling user behavioral tendencies. A balanced
Malay–English cyberbullying corpus comprising 52,140 annotated instances was
constructed through multi-source dataset integration and standardized labeling
to support model training and evaluation. The proposed architecture integrates
TF-IDF (Term Frequency–Inverse Document Frequency) lexical representations with
CNN-based contextual feature learning, enabling the model to capture both
surface-level linguistic cues and deeper semantic patterns characteristic of
code-mixed discourse. Experimental evaluation demonstrates strong predictive
performance, achieving 98.42% classification accuracy with consistently high
precision, recall, and F1-scores across severity categories. Ablation analysis
further shows that neither lexical nor neural representations alone sufficiently
capture cyberbullying severity, whereas their hybrid integration yields more
balanced classification outcomes. Beyond predictive performance, a rule-based
behavioral profiling module enhances interpretability by mapping severity
predictions to distinct interaction archetypes. These findings demonstrate that
combining hybrid deep-learning architectures with interpretable behavioral
analysis provides a scalable and context-sensitive approach for cyberbullying
severity detection in multilingual code-mixed environments, supporting more
effective automated moderation and digital safety interventions. |
|
Keywords: |
Convolutional Neural Networks, Cyberbullying Detection, Malay–English
Code-Mixing, Natural Language Processing, Severity Classification, User Behavior
Profiling |
|
DOI: |
|
|
Source: |
Journal of Theoretical and Applied Information Technology
15th June 2026 -- Vol. 104. No. 11-- 2026 |
|
Full
Text |
|
|
|
