Asian Journal of Research in Computer Science
Not a member yet
792 research outputs found
Sort by
Energy-aware Multi-agent RAG Planner for Edge Devices Using vLLM and Model Pruning
No full textIt looks into addressing the growing need to minimize energy use in systems that apply retrieval with language generation. As the use of large language models (LLMs) increases, their energy and operational costs go up, so there is a real need to find energy-saving ways to work with them. Even with the development of RAG architectures, energy efficiency is frequently neglected, causing bigger computational requirements.
To bridge this research gap, we expect an energy-aware RAG planning approach to make use of vLLM which is a highly effective and optimized language model serving system and also rely on model pruning strategies. Minimizing the energy used is the main goal, not deteriorating the accuracy and quality of the answers retrieved. Because our system uses lightweight serving from vLLM and removes unnecessary parameters in the language models, it finds a good balance between workload and quality of results.
Among our findings are (1) a RAG planning algorithm that adjusts the necessary model complexity to match query requirements, (2) model pruning techniques developed for energy-saving purposes in RAG and (3) thorough testing that confirmed significant energy cost reductions—up to 40%—with similar prediction performance. This means there is a good chance making RAG models more sustainable could be possible by using both improved serving frameworks and reducing model size. This allows for AI applications to be built considering energy efficiency in any place where resources are scarce
AI-Governed Security Frameworks for Virtualized Enterprises: Preventing Data Breaches and Ensuring Compliance
No full textThis study investigates the role of artificial intelligence AI–governed security frameworks in addressing cybersecurity risks, preventing data breaches, and ensuring regulatory compliance in virtualized enterprises. Virtualization technologies have enhanced scalability and efficiency but also introduced complex vulnerabilities such as misconfigurations, insider threats, and compliance gaps that traditional perimeter-based security models cannot adequately address. The objectives of this research are to identify the dominant sources of risk in virtualized environments, evaluate the performance of AI-driven detection models against traditional methods, and examine the capacity of AI systems to automate compliance with sectoral and international regulations. The analysis, conducted using multiple open-source datasets, revealed that misconfigurations, particularly unencrypted storage, overly permissive identity and access management (IAM) roles, and open ports collectively accounted for more than 70 percent of recurring vulnerabilities. A supervised Random Forest model demonstrated superior performance in detecting malicious network activity, achieving an accuracy of 96.4 percent and significantly outperforming a rule-based baseline (p < 0.001). Compliance mapping indicated that frameworks such as HIPAA and GDPR achieved 80 percent coverage of high-severity vulnerabilities, though remediation times varied considerably, reflecting inefficiencies in operational response. The findings underscore that while AI substantially enhances detection accuracy and strengthens compliance oversight, unresolved challenges remain, including inconsistent remediation speed, dataset bias, and risks of privilege escalation. To address these gaps, the study recommends hybrid AI–human oversight protocols, standardized remediation workflows, pre-deployment bias audits, and sector-specific benchmarks. These contributions provide actionable insights for advancing secure, efficient, and trustworthy governance of virtualized enterprise systems
Regulating AI in Cybersecurity: Balancing Challenges, Opportunities, and Global Frameworks
No full textArtificial Intelligence (AI) is simultaneously a transformative enabler and a regulatory challenge in the fast-evolving domain of cybersecurity. Artificial intelligence is gradually taking over human labour and decision-making. Numerous industries, including manufacturing, transportation, logistics, government, health, education, and commerce, have made substantial use of it. This study investigates how AI is currently deployed within cybersecurity frameworks, the ethical and legal risks it introduces, and the adequacy of emerging regulatory responses. Methodologically, a systematic literature review aligned with PRISMA principles was applied, drawing from peer-reviewed articles, regulatory reports, and international policy guidelines. Key findings show that while AI improves detection, automation, and cyber resilience, it also generates risks of algorithmic bias, data privacy breaches, and malicious misuse. Comparative analysis of the EU AI Act, the US NIST AI Risk Management Framework, China’s AI regulations, and OECD principles reveals fragmented governance. This paper contributes by proposing the AI-Cybersecurity Regulatory Integration Model, a conceptual framework aligning risk management, accountability, and cybersecurity safeguards. We conclude that effective governance requires harmonisation of global standards, operationalisation of ethical principles, and sustained cross-sector collaboration
Blooming Insights: Flower Classification Using Data Mining Techniques
No full textPrecise flower classification is essential in botanical research, agricultural diagnostics, and biodiversity monitoring, as accurate species identification underpins ecological sustainability and informed decision-making. This study rigorously evaluates the utilization of data mining techniques in flower classification, specifically contrasting conventional algorithms like decision trees, K-means clustering, and support vector machines with contemporary deep learning methods, particularly convolutional neural networks (CNNs) augmented by transfer learning and hybrid feature strategies. A systematic literature review examined peer-reviewed research published from 2020 to 2025. The results indicate that although conventional data mining models provide computational efficiency and interpretability, they frequently exhibit suboptimal performance on high-dimensional image datasets. CNN-based architectures consistently exhibit enhanced accuracy, robustness, and scalability, especially when integrated with data augmentation and optimization methods. Nonetheless, significant limitations persist, including constrained generalizability across varying environmental conditions, insufficient explainability, and difficulties in real-time implementation. This review enhances theoretical understanding and practical implementation by examining the progression of flower classification methodologies and suggesting future research avenues that emphasize balanced models integrating performance, transparency, and adaptability in field applications
Enhancing Security Practices across the Software Development Lifecycle: The Role of Artificial Intelligence
No full textAims: This study examines the role of Artificial Intelligence (AI) in enhancing software security across the Software Development Lifecycle (SDLC). It seeks to highlight how AI-driven tools and techniques strengthen security practices, improve efficiency, and address evolving cyber threats.
Study Design: The paper adopts a conceptual review approach, providing an overview of prominent AI methods such as machine learning (ML), deep learning (DL), and natural language processing (NLP). These are analyzed in relation to their applications in different phases of the SDLC.
Methodology: The review explores AI integration into five stages of the SDLC: requirements, design, implementation, testing, and deployment/maintenance. It evaluates applications such as predictive risk analysis, secure design recommendations, vulnerability scanning, automated code review, penetration testing, anomaly detection, and real-time security monitoring.
Results: Findings show that AI improves predictive threat modelling in the requirements stage and assists with secure design and attack surface analysis during design. In implementation, AI-based tools such as SonarQube and Checkmarx enhance code quality and detect security flaws. In testing, AI enables automated penetration testing and test case generation. At deployment and maintenance, AI supports real-time monitoring and predictive maintenance. Overall, AI adoption in the SDLC enhances efficiency, reduces human error, and accelerates threat identification. However, challenges include integration complexity, false positives, and data privacy concerns.
Conclusion: AI significantly strengthens security throughout the SDLC by providing adaptive, automated, and proactive defenses. While integration challenges remain, its benefits outweigh limitations, making AI-driven practices essential for robust and agile software security in the present and future
Dark Data in Digital Health: A Predictive Framework for Identifying and Utilizing Underreported Clinical Signals
No full textThis study developed a predictive framework to uncover underreported clinical signals hidden within dark data in digital health systems, addressing the paradox of abundant data but limited insights. It highlighted the importance of dark data, unanalyzed clinical records and the potential of AI to reveal hidden patterns while maintaining ethical standards. The research reviewed homomorphic encryption and AI integration, identifying a lack of real-time analysis in telehealth. Using a CRISP-DM-based methodology, machine learning was applied to datasets such as MIMIC-IV and PhysioNet, with preprocessing techniques like KNN imputation and DBSCAN outlier detection. Results showed neural networks achieved a 94% AUC-ROC, detected 32 new clinical signals, and improved rare disease identification by 50%. Ethical anonymization maintained 97% data utility, though dependence on historical data was a limitation. The novelty of this framework lies in merging dark data analytics with secure AI to enhance healthcare decision-making, patient safety, and precision medicine. Future work recommends real-time data integration, explainable AI, and standardized ethical protocols for scalability
Design and Implementation of Online Crime Report System using Rapid Application Development (RAD) Methodology
No full textIn recent years, online crime reporting systems have become increasingly important for law enforcement and community safety. However, many existing systems lack features that enhance usability and accessibility for diverse populations. This paper provides a comprehensive overview of the design and implementation process for an Online Crime Report System, employing the Rapid Application Development (RAD) methodology. Our proposed system integrates advanced data analytics, a multi-language interface, and anonymous reporting capabilities, which address these limitations and provide significant improvements over current solutions. Additionally, the system is designed with scalability, high availability, low latency, reliability, and durability in mind to ensure it can effectively support crime reporting at scale in real-world scenarios. It explores the technical aspects of system development, discusses the challenges encountered, and the solutions devised, and evaluates the effectiveness of the RAD approach in this context. By examining this case study, we aim to contribute to the growing body of knowledge on the application of agile development methodologies in law enforcement technology and to provide insights for future implementations of similar systems
The Role of Artificial Intelligence in Banking and Fraud Prevention: A Cross Sectional Study in Ghana
No full textIntroduction: The increasing integration of Artificial Intelligence (AI) in the banking sector has reshaped traditional financial services, particularly in the context of fraud prevention. This cross-sectional study in Ghana aimed to investigate the current state and perceived effectiveness of AI applications in banking, focusing on its role in fraud prevention.
Methods: The research data was acquired through interviews and surveys conducted with customers and bank officials. A total of 363 participants took part in the survey, comprising 200 customers and 163 staff members selected from five banks in Ghana. Structured questionnaires were distributed electronically and in print to gather quantitative and qualitative data.
Results: The findings reveal a significant level of awareness (70.0%), understanding (75.0%) and 62.0% experience with AI in the banking sector among the participants. An overwhelming 88.0% express a preference for AI-based support over human-based support. About 97.2% believe that AI systems prioritize robust privacy measures influencing their perception of AI in fraud prevention. Furthermore, 87.5% perceive AI systems as consistently providing precise and reliable results, enhancing their confidence in the technology. The perception of AI\u27s effectiveness in fraud prevention is closely tied to its capacity to adapt to new and emerging fraud tactics, with 66.6% emphasizing the importance of this adaptability.
Conclusion: These findings contribute to understanding the nuanced perspectives of users in Ghana regarding AI in the banking sector, providing insights for financial institutions, policymakers, and educators aiming to enhance AI adoption and trust
Modular Co-attention Networks in Nepali Visual Question Answering Systems
No full textVisual question answering (VQA) has been regarded as a challenging task requiring a perfect blend of computer vision and natural language processing. As no dataset was available to train such a model for the Nepali language, a new dataset was developed during the research by translating the VQAv2 dataset. Then the dataset consisting of 202,577 images and 886,560 questions was used to train an attention-based VQA model. The dataset consists of yes/no, counting, and other questions with primarily one-word answers. Modular Co-attention Network (MCAN) was applied to the visual features extracted using the Faster RCNN framework and question embeddings extracted using the Nepali GloVe model. After co-attending the visual and language features for a few cascaded MCAN layers, the features are fused to train the whole network. During evaluation, an overall accuracy of 69.87% was obtained with 81.09% accuracy in yes/no type questions. The results surpassed the performance of models developed for Hindi and Bengali languages. Overall, novel research has been done in the Nepali Language VQA domain paving the way for further advancements
Sarcasm Detection in Pidgin Tweets Using Machine Learning Techniques
No full textDetecting sarcasm in social media is of growing importance for applications such as monitoring, consumer feedback, and sentiment analysis. However, detecting sarcasm in Pidgin tweets poses unique challenges due to the blend of English and Pidgin languages, along with local cultural references. Existing models for sarcasm detection in English lack appropriate annotated data for Pidgin. This scarcity hinders the development of effective machine learning models. This research aims to address these challenges and create a model for accurate sarcasm detection in Pidgin tweets. Logistic Regression, XGBoost, Random Forest, and Vanilla Artificial Neural Network (ANN) classifiers were assessed, focusing on accuracy, precision, recall, and F1-score metrics on sarcasm data collected by curating and pre-processing a dataset of Nigerian Pidgin tweets. The XGBoost model demonstrated notable performance, attaining an accuracy of 85.78%, precision of 88.57%, recall of 94.44%, and F1-score of 91.41%. These outcomes underscored the model\u27s prowess in discerning sarcastic and non-sarcastic expressions. By unfolding the intricacies of language in the Nigerian context, this research into sarcasm identification in Nigerian Pidgin text data introduced a comprehensive pipeline encompassing data curation, exploratory analysis, culturally tailored pre-processing, model training, evaluation, and prediction