Asian Journal of Research in Computer Science

Artificial Intelligence and Post-Quantum Cryptography for Healthcare Cybersecurity: A Critical Review of Techniques, Architectures, and Open Challenges

2026-09-26T12:18:03+00:00

Aims: This critical review examines how artificial intelligence (AI) and post-quantum cryptography (PQC) can be combined to strengthen cybersecurity in healthcare information systems, with an emphasis on the Internet of Medical Things (IoMT), clinical data platforms, zero-trust access control and cryptographic migration.

Study Design: The review uses a structured literature-search and evidence-mapping approach. It emphasises peer-reviewed sources in computer science, cybersecurity, healthcare informatics and cryptography, as well as authoritative standards.

Methodology: Evidence was organised into healthcare/IoMT cybersecurity, AI-enabled detection and response, PQC and migration, and integrated security architecture. NIST, FDA and IETF material was used to distinguish finalised standards from emerging work. Studies were compared by security function, deployment layer, resource overhead, interoperability, clinical risk and empirical validation.

Results: AI and PQC address different but complementary requirements. AI supports anomaly detection, behavioural analytics, alert prioritisation and migration assistance, but is exposed to evasion, poisoning, drift and explainability problems. NIST FIPS 203, FIPS 204 and FIPS 205 provide finalised standards for ML-KEM, ML-DSA and SLH-DSA. Major deployment barriers include constrained devices, larger keys and signatures, legacy dependencies, interoperability, implementation leakage, regulation and limited real-world validation. A five-layer reference architecture and a phased migration architecture are proposed.

Conclusion: Healthcare organisations should begin cryptographic discovery and risk-based PQC migration while strengthening AI security controls. Future work should prioritise hardware-aware PQC benchmarks, adversarially robust AI, clinically safe automated responses, cryptographic agility and multi-site validation.

Power-aware Lightweight Authentication in Resource-constrained Devices: A Review

2026-10-03T09:10:15+00:00

The use of Internet of Things (IoT) applications has increased. However, IoT devices have limited computing, memory, bandwidth, and battery power. Therefore, authentication methods that provide security while using fewer resources are required. Traditional security mechanisms, such as Public Key Infrastructure (PKI) and Identity-Based Encryption (IBE), have high computational and communication overheads, making them infeasible for low-end IoT devices. This review examines lightweight authentication methods for resource-limited IoT devices. It compares symmetric-key-based, hash-based, lightweight cryptography-based, Physical Unclonable Function (PUF)-based, multifactor, and hybrid methods. The analysis considered computational and communication costs, memory requirements, hardware demands, energy use, and security strength. The findings show that hash- and PUF-based methods offer a balance between security and resource use. Hybrid methods combining PUFs with lightweight techniques, such as hash functions, XOR operations, nonces, and ASCON, operate with low computational, communication, and power demands while providing mutual authentication, protection against replay attacks, confidentiality, information integrity, and secure session-key generation. The work also describes challenges, including key distribution, scalability, protection against physical attacks, balancing security and power consumption, and the lack of standardised frameworks. Overall, PUF-based lightweight hybrid approaches have the potential to support secure, low-power authentication systems in future IoT and 6G environments.

Artificial Intelligence Classification Algorithms for Leukaemia Diagnosis: A Structured Review of Model Performance and Clinical Readiness

2026-10-03T11:07:58+00:00

Leukaemia diagnosis and classification increasingly combine morphology with immunophenotypic, cytogenetic and molecular information, while artificial intelligence (AI) offers a potential means of standardising image-based assessment and reducing repetitive workload. This structured review critically examines contemporary AI approaches for leukaemia cell and case classification, with emphasis on dataset scale, unit of analysis, validation design, reported performance and clinical readiness. Peer-reviewed literature was identified through a targeted PubMed-focused search updated on 30 July 2026, with publisher and DOI records used to verify study design, numerical results and bibliographic identity. Representative primary human studies were selected only when the reported task, unit of analysis, testing design and principal performance metrics could be verified from the primary report. The evidence shows substantial technical progress, but the interpretation of apparently high performance changes markedly with the level at which generalisation is tested. Image-wise or closely related internal partitions provide weaker evidence of transportability than patient-level internal testing, while genuinely independent institutional or imaging-platform evaluation provides the most clinically informative test of robustness. Across acute lymphoblastic leukaemia, acute myeloid leukaemia and acute promyelocytic leukaemia applications, patient-level separation, external testing, class-balance reporting, transparent reference standards and multimodal diagnostic integration remain decisive. AI should therefore be considered an adjunct to expert haematopathology rather than a stand-alone substitute for contemporary integrated diagnosis. Multicentre prospective evaluation, rigorous reporting and clinically meaningful human-AI workflow studies are priorities for translation.

Explainable Artificial Intelligence for Zero-Day Cyberattack Detection in Healthcare: A Critical Review and an Integrated Framework for Clinical Cybersecurity

2026-10-05T12:09:43+00:00

Healthcare delivery organisations have become preferred targets for cyber intrusion, and the clinical consequences of successful attacks now extend beyond information loss to measurable disruption of emergency, critical and diagnostic care. Signature-based defences cannot recognise previously unseen exploits, which has driven interest in machine learning detectors capable of identifying anomalous behaviour without prior knowledge of an attack pattern. Because such detectors are opaque, explainable artificial intelligence has been proposed as the mechanism through which their outputs can be audited, trusted and acted upon in clinical settings. This review examines whether that proposition is supported by the available evidence. Literature was identified through structured searching of open scholarly indexes, supplemented by backward and forward citation searching, and appraised for methodological adequacy, evidential strength and relevance to clinical deployment rather than for benchmark performance alone. The synthesis indicates that the two research streams have developed with limited contact. Work on detection of previously unseen attacks is dominated by retrospective evaluation on a small number of network datasets in which novelty is simulated by withholding labelled classes, a design that systematically flatters reported performance. Work on explanation is dominated by post-hoc feature attribution, whose faithfulness, stability and adversarial robustness remain contested, and which is poorly suited to characterising events that lie outside the training distribution. Healthcare-specific studies inherit both limitations while adding constraints related to medical device heterogeneity, data governance and clinical safety. Evidence that explanation improves analyst or clinician decision quality in security contexts is scarce, and current regulatory expectations for transparency are not matched by validated evaluation methods. An integrated framework is proposed that separates detection novelty, explanation target, explainee role and governance obligation as distinct design decisions, and that treats explanation quality as an empirical property requiring prospective human-centred evaluation rather than an assumed benefit.

Artificial Intelligence, Cybersecurity and Digital Learning Resilience in Higher Education: An Integrative Review and Capacity-Sensitive Framework for Resource-Constrained Universities

2026-10-17T09:43:54+00:00

Purpose: Artificial intelligence (AI) is now embedded in teaching, assessment and administration in higher education, simultaneously widening educational opportunity and institutional cyber exposure. This review examines how AI-related and conventional cyber threats affect the continuity of digital learning, how AI can both amplify and mitigate those threats, and which integrated responses are feasible for resource-constrained universities, with Nigeria as the illustrative context.

Design/Methodology/Approach: A narrative integrative review following established integrative-review guidance synthesised 60 sources published mainly between 2019 and 2026, comprising peer-reviewed studies, technical standards, policy and legal instruments, and institutional or industry reports. Sources were organised by threat mechanism, institutional vulnerability, educational consequence, response and implementation constraint, and each was classified by the type of inference it can support.

Findings: Seven overlapping threat clusters were identified: AI-enhanced social engineering and identity abuse; ransomware and extortion; privacy leakage and surveillance; platform and supply-chain compromise; manipulation of AI-enabled learning systems; synthetic media and assessment-integrity attacks; and human, governance and infrastructure weaknesses. Experimental evidence shows that large language models lower the cost of persuasive phishing, while AI-text detectors and automated proctoring raise reliability and fairness concerns. AI-assisted defence offers measurable triage benefits but introduces false alarms, concept drift, privacy exposure and automation bias.

Originality/Value: The proposed AI-Cybersecurity-Digital Learning Resilience (AIC-DLR) Framework adapts NIST CSF 2.0 functions to the educational mission, overlays NIST AI risk-management guidance, and treats usable, equitable recovery of teaching and assessment, rather than system restoration alone, as the protected outcome. A tiered roadmap and a set of operationally defined indicators translate the framework into testable practice.

Practical Implications: Priorities for low-capacity institutions are accountable ownership, phishing-resistant identity for privileged users, tested and isolated backups, proportionate data practices, governed AI use and pre-authorised alternative learning channels. The framework is conceptual and unvalidated; comparative implementation studies are required to establish its effect on recovery time and equitable access.

A Multi-Layer Security Evaluation Framework for Hardening Kubernetes in Devsecops Pipelines: A Quantitative Experimental Evaluation

2026-09-26T06:44:58+00:00

Background: Kubernetes is widely used to orchestrate containerised workloads in DevSecOps environments; however, its complex configuration, access-control mechanisms, software supply chain, and runtime operations create multiple security challenges. Existing security controls often address these risks independently, creating a need for an integrated and measurable framework that evaluates prevention, software-supply-chain integrity, and runtime detection and response within a unified Kubernetes environment.

Aims: To design, implement, and quantitatively evaluate a Multi-Layer Security Evaluation Framework (MSEF) that hardens Kubernetes workloads across prevention, software-supply-chain integrity, and runtime detection and response.

Study Design: A controlled quantitative experimental design using comparative baseline and hardened Kubernetes environments.

Place and Duration of Study: A cloud-based experimental testbed implemented on Google Kubernetes Engine (GKE) and operated through a DevSecOps toolchain during the 2026 evaluation period.

Methodology: The Prevention Layer used OPA Gatekeeper and Kubernetes Pod Security Admission to measure Manifest Blocking Rate (MBR), Network Policy Enforcement Rate (NPER), and Secrets Management Enforcement Rate (SMER); the Integrity Layer used Kyverno and Sigstore Cosign to measure Secrets Management Enforcement Rate (SMER) and Signature Policy Rejection Rate (SPR); and the Detection Layer used Falco, Falcosidekick, and a Kubernetes Runtime Evaluation (KRE) handler to evaluate Runtime Detection Rate (RDR), Mean Time to Detect (MTTD), Operational False Positive Rate (OFPR), and Runtime Response Success Rate (RRSR). Layer-level effectiveness composite scores were defined for prevention, integrity, and detection. Terraform set up and configured the cloud infrastructure, ArgoCD reconciled platform services and workloads, and GitHub Actions deployed the repeatable evaluation.

Results: The final validated prevention experiments achieved MBR = 1.00, SMER = 1.00, and NPER = 1.00; the integrity experiment achieved SPR = 1.00; and, in the detection experiment, Falco detected all five evaluated runtime attack windows (RDR = 1.00). The measured MTTD was 23.00 seconds, and no operational false-positive windows (OFPR = 0.00) were observed in the five benign workload windows. The initial MBR of 0.95 exposed a secret-access RBAC policy gap that was subsequently corrected. The recorded RRSR run produced 0.78 because Falco event-to-KRE response-rule alignment was still under validation; this value is therefore treated as an integration result rather than evidence of successful automated remediation. Layer-level effectiveness composite scores for prevention and integrity are 100%, while detection is 90%.

Conclusion: MSEF's unique contribution to knowledge is the multi-layer integration of controls with explicit experiments, quantitative metrics, feedback-driven policy refinement, and browser-based reporting, rather than the deployment of individual security tools, because it provides a replicable and measurable defence-in-depth model for Kubernetes security.

Self-healing Microservice Architecture Using Autonomous AI Agents for Cloud Applications

2026-09-28T13:37:41+00:00

Cloud-native microservice applications provide scalability and deployment flexibility but remain susceptible to service failures, cascading faults, and delayed recovery. This study presents a Self-Healing Microservice Architecture (SHMA) that combines Kubernetes-based container orchestration, autonomous AI agents, a binary failure-prediction component, and reinforcement-learning-based recovery selection. The prototype was implemented with Spring Boot, Docker, Kubernetes, TensorFlow, Prometheus, Grafana, and PostgreSQL. Evaluation considered Mean Time to Recovery (MTTR), availability, recovery accuracy, response time, throughput, transaction success, and failure-prediction metrics. In the reported controlled testbed, MTTR decreased from 240 s to 48 s, availability increased from 98.2% to 99.8%, recovery accuracy increased from 84% to 98%, and service downtime decreased from 12 min to 2 min. The reported 1,000-observation confusion matrix yielded 95.7% accuracy, 96.5% precision, 95.2% recall, and 95.8% F1-score for failure prediction. In the increasing-workload experiment, response time remained at or below 1.18 s through 1,200 concurrent users, peak observed throughput reached 1,620 requests/s, and the successful transaction rate was 99.9%. These findings provide descriptive evidence that combining predictive monitoring with policy-driven recovery can improve fault-management responsiveness in a controlled Kubernetes environment. The results are interpreted within the limits of a single-cluster evaluation and aggregate point estimates, and broader generalisation requires repeated and multi-environment validation.

A Retrieval-augmented Generation Framework for Policy-grounded Course Advising in Higher Education

2026-10-01T10:08:22+00:00

Course advising in higher education requires students and advisers to integrate information from course catalogues, programme requirements, departmental handbooks, prerequisite rules, and institutional regulations. This study presents a Retrieval-Augmented Generation (RAG) framework for policy-grounded undergraduate course advising at Babcock University, Nigeria. The prototype combines a structured academic database, an indexed institutional knowledge base, a large language model, and a separate validation layer for prerequisite and credit-load constraints. The implementation includes student-profile management, natural-language querying, knowledge-base administration, authentication, and chat-log review. Evaluation used 18 institutional source documents segmented into 246 indexed chunks, 30 simulated student profiles, 60 advisory queries, and benchmark judgements from three academic advisers. The qualitative beta evaluation examined policy retrieval, policy grounding, prerequisite and credit-limit validation, response clarity, ambiguity handling, advising usefulness, and observed failure conditions. The findings provide formative evidence that the framework can retrieve relevant institutional information, incorporate evidence into advisory responses, and apply deterministic checks to explicit academic rules. However, context fragmentation and ambiguity in source documents remained important limitations, and the evaluation did not provide statistical performance estimates. The framework is therefore positioned as a decision-support tool that retains human oversight for ambiguous or high-stakes advising cases and requires larger controlled evaluation before measurable advising performance can be established.

AI-Powered Auto-healing Agent for Real-time DataOps Pipeline Failures

2026-10-03T10:55:55+00:00

Modern DataOps environments are increasingly complex because of distributed continuous integration/continuous delivery (CI/CD) frameworks within lakehouse architectures. This scale amplifies failures arising from schema drift, transient network errors, and upstream data inconsistencies, resulting in prolonged downtime and increased engineering overhead. This study presents an auto-healing agent for real-time incident detection, root cause analysis (RCA), and transactionally safe remediation of routine failures. For complex cases, structured escalation preserves timely human oversight. The framework integrates open lakehouse technologies into a unified system: Apache Hudi supports near-real-time streaming signals; Apache Iceberg provides consistent time-travel snapshots for state comparison; and Delta Lake ensures safe ingestion and recovery without disruption. Detection employs an imbalance-aware weighted Random Forest classifier with class and sample weighting and a modified Gini impurity objective to address extreme class imbalance (~97.4%) in operational logs. On the Hadoop Distributed File System (HDFS) anomaly benchmark (>11 million logs; 37:1 imbalance), the model achieved a weighted F1-score of 85.74%, covering 468,279 detected anomalies. The policy prioritises anomaly recall (37.59%) over precision (4.61%) to reduce missed detections, directing 99.8% of cases to Level-2 escalation. Healing actions respect dependency paths through confidence-based checks before remediation to control risk propagation. Hybrid RCA combines rule-based analysis with simulated generative reasoning without relying on production LLMs. Operational key performance indicators (KPIs), including recovery time, indicate a 20–30% reduction in repetitive Level-1 tickets and a measurable improvement in mean time to recovery (MTTR) for automatable incidents. The results support a scalable, transparent, and safety-first framework for reliable DataOps pipeline operations.

Password Management Practices, Authentication Burden, and Institutional Cybersecurity Support among Students and Staff at a Health Training Institution in Ghana: A Cross-Sectional Study

2026-10-03T14:35:26+00:00

Background: Secure authentication is increasingly important in health-professions education because students and staff use digital platforms that contain personal, academic, and potentially practice-related information. Yet password security is constrained by memory burden, password reuse, limited uptake of protective tools, and institutional policies that may prioritise complexity over usability. Evidence from specialised health-training institutions in sub-Saharan Africa remains scarce.

Objective: The study aims to assess password construction, reuse, protective authentication practices, institutional cybersecurity support, and password-related access burden among computer users at Tepa Nursing and Midwifery Training College, Ghana.

Methods: A descriptive cross-sectional survey was conducted in 2025 among 320 students and staff. A structured online questionnaire captured demographic characteristics, password practices, use of password-management and breach-detection tools, multifactor authentication, institutional training, and authentication-related access difficulties. Data were analysed in IBM SPSS Statistics version 25 using frequencies and percentages.

Results: Most respondents were students (94.7%), aged 18–24 years (80.0%), and daily computer users (55.0%). Although 57.8% used alphanumeric passwords, 57.8% incorporated personal information, 48.4% used passwords shorter than eight characters, and 55.9% used predictable patterns. Eighty per cent reused passwords at least sometimes. Only 37.8% used multifactor authentication, 15.9% used breach-detection tools, and 30.9% reported institutional cybersecurity training. Password burden was substantial: 61.2% had difficulty tracking multiple passwords and 59.7% reported frustration with complex requirements.

Conclusion: Frequent digital use did not translate into secure authentication behaviour. The combination of reuse, predictable credentials, low uptake of protective controls, limited training, and high password burden indicates a need for a user-centred institutional authentication programme. Priority measures include long unique passwords or passphrases, compromised-password screening, approved password managers, phased multifactor authentication, usable recovery pathways, and recurring practical cybersecurity education.

Comparative Analysis of Feature Extraction Methods on Convolutional Neural Network Models for Classification of Healthy and Unhealthy Cacao Pods

2026-10-14T10:45:04+00:00

Objective classification of cacao pods can support consistent assessment of pod health under field conditions. This study developed and compared convolutional neural network (CNN) models for classifying healthy and unhealthy cacao pods using raw images and three feature extraction approaches: Local Binary Pattern (LBP), Edge Feature (EF), and Gabor Filter (GF). A total of 500 cacao pod images were collected using a high-resolution DSLR camera and labelled by expert cacao farmers, comprising 250 healthy and 250 unhealthy pods. Images were preprocessed, and the datasets were partitioned into 70% training, 20% validation, and 10% test sets. Four CNN configurations were evaluated: a baseline CNN using raw images and CNN models using LBP-, EF-, and GF-processed inputs. The models were trained for 100 epochs and evaluated using confusion-matrix-derived classification metrics. The CNN with Edge Feature preprocessing achieved the highest reported performance, with 96.15% accuracy, a Kappa score of 0.923, and an F-measure of 96.15%. The LBP and Gabor Filter configurations achieved accuracies of 90.38% and 88.46%, respectively, while the baseline CNN achieved 69.23%. The findings indicate that feature extraction can improve CNN-based cacao pod classification, with edge-based preprocessing providing the strongest performance among the configurations evaluated. Further validation across broader field conditions is warranted before deployment.

From Fragmented Protective Practices to Secure Digital Learning: Developing the Secure-Dl Cybersecurity Adoption Framework for Higher Education

2026-10-14T10:50:34+00:00

Digital learning depends on students consistently applying several layers of cybersecurity protection, yet higher-education interventions often rely on general awareness messages or isolated technical training. This study developed an evidence-informed, theory-integrated framework for moving students from fragmented protective practices towards secure digital learning. Primary survey data were obtained from 393 Computer Science undergraduates in universities in Enugu State, Nigeria, using the 60-item Utilisation of Cybersecurity Applications for Digital Learning Questionnaire. Five domains were assessed: antivirus, firewall and intrusion detection/prevention, encryption, multi-factor authentication (MFA), and virtual private network (VPN) use. Domain means were benchmarked against the prespecified functional criterion of 2.50 using one-sample tests and standardised mean differences. Empirical gaps were then mapped to mechanisms from Social Cognitive Theory, the Unified Theory of Acceptance and Use of Technology, and Protection Motivation Theory, and aligned with multilevel implementation principles. All five domains were below the functional benchmark. Antivirus use was closest to the criterion (M=2.36, SD=0.57, d=-0.24), whereas firewall (M=2.07, d=-0.59), encryption (M=2.08, d=-0.60), MFA (M=2.12, d=-0.58), and VPN use (M=2.10, d=-0.61) showed moderate deficits. The resulting SECURE-DL framework comprises six linked functions: Screen and segment; Enable secure defaults; Cultivate capability; Use threat-and-coping messages; Reinforce norms and feedback; and Evaluate and adapt. It specifies actions at the student, course/programme, and institutional levels, together with process, behavioural, and resilience indicators. The framework shifts cybersecurity education from awareness-only provision towards supported, repeated, and measurable protective practice. Prospective testing is required before causal effectiveness or generalisability can be established.