Accurate estimation of software defects is essential in building reliable and cost-efficient systems, especially when predictions must be made at an early stage of the Software Development Life Cycle (SDLC). This paper presents a human-in-the-loop causal estimation model that incorporates Bayesian Belief Networks (BBNs) and fuzzy logic to represent uncertainty and establish relationships among top-ranked metrics from the requirements, design, and coding phases. In contrast to other hybrid BBN–fuzzy approaches, the proposed architecture directly combines expert judgment, novice assessments, and probabilistic inference, facilitating effective defect estimation in environments where skilled experts are in short supply. The model was evaluated using a sample of 31 software projects from prior work to assess early-phase causal defect estimation under data-sparse conditions. Statistical evaluation using Mean Magnitude of Relative Error (MMRE), Balanced Mean Magnitude of Relative Error (BMMRE), Mean Absolute Error (MAE), Root Mean Squared Error (RMSE), and nonparametric significance testing demonstrates that the proposed approach achieves improved estimation accuracy relative to benchmark methods. The Wilcoxon signed-rank test indicates a statistically significant reduction in estimation error for the proposed framework compared to the benchmark methods. Notably, novice estimators, when guided by the causal model, generated estimates closely aligned with expert assessments, highlighting the usefulness of the framework in agile and resource-constrained development settings.

Natural disasters have a major effect on the Earth and cause severe repercussions. The use of social networks is a valuable method for gathering information and insight during such situations. Natural language processing techniques that use deep learning and machine learning models show strong potential for categorising disaster-related data. To enhance the classification and recognition of social media information related to natural and man-made disasters, this research introduces a novel approach, DisasterSense. Many existing detection tools misclassify man-made disasters as natural because social media texts often lack clear linguistic patterns. In this study, Dictionary-Driven Named Entity Recognition was applied to create a word-based disaster dictionary for categorising natural and man-made disasters, which was then combined with pattern mining to identify recurrent themes, patterns, and relationships among frequently used phrases. Using frequent patterns extracted with the Rapid Automatic Keyword Extraction (RAKE) model, an unsupervised technique for identifying keywords and phrases from text, a new pattern-based disaster dictionary was developed to enhance classification accuracy through pattern comparison. The effectiveness of the proposed approach was evaluated by comparing it with supervised and unsupervised models. This research aims to help emergency response organisations respond more rapidly and mitigate the impacts of disasters. Experiments on disaster-related data demonstrated that our approach achieved the highest accuracy of 0.789, outperforming all other well-known classification methods evaluated in this study.

Artificial intelligence is reshaping hospital care delivery through federated learning pipelines, edge-cloud inference architectures, and AI-driven clinical decision support. Yet the translation of these AI capabilities into patient-centred, institutionally governable, and humanised hospital systems remains fragmented across the literature. This paper addresses that gap through a PRISMA-compliant systematic evidence synthesis of 116 included studies and reports (inter-rater reliability κ = 0.884 , 95

Federated learning (FL) has emerged as an effective paradigm for collaborative model training in Industrial Internet of Things (IIoT) environments by enabling distributed devices to learn shared models without exchanging raw data. However, existing FL defence mechanisms predominantly rely on either behavioural analysis of client updates or semantic validation of model performance, limiting their ability to detect sophisticated poisoning and stealthy backdoor attacks that evade single-dimensional trust assessment. This paper proposes DBST-FL, a dynamic behavioural and semantic trust framework for robust federated learning in the Industrial IoT. The proposed framework evaluates each client through two complementary trust dimensions: a behavioural trust layer that measures gradient alignment, historical consistency, and collective deviation and a semantic trust layer that assesses benign utility and template-free semantic stress validation using server-side data. The two trust scores are integrated through a non-compensatory multiplicative trust fusion mechanism, ensuring that weaknesses in one trust dimension cannot be masked by strengths in the other. The resulting trust score guides a trust-aware aggregation strategy that reduces the influence of malicious participants while preserving the contributions of reliable clients. Extensive experiments are conducted on the Edge-IIoTset and UNSW-NB15 datasets using ANN, 1D-CNN, and LSTM models under multiple poisoning and backdoor attack scenarios. The proposed framework achieves overall classification performance competitive with the strongest robust aggregation baselines while consistently delivering stronger resilience against adversarial attacks and lower backdoor attack success rates than representative trust-based and Byzantine-robust aggregation methods, all while maintaining linear per-round computational complexity suitable for large-scale IIoT deployments. The results demonstrate that integrating behavioural and semantic trust within a unified aggregation framework provides an effective and scalable defence against advanced adversarial threats in federated learning.

This study presents a federated learning (FL) framework for predictive maintenance (PdM) of satellite battery systems, integrating the NASA BP930 dataset with MACCOR-generated simulations to create a hybrid training environment that reflects both real and synthetic operating conditions. Unlike previous FL studies in PdM, the proposed approach models resource constraints such as power variability and communication limits, replicating conditions in large-scale industrial and aerospace networks. Feedforward neural networks (FFNNs) are deployed at edge nodes, and global aggregation is performed through a reliability-weighted FedAvg algorithm. Experimental results show that our proposed framework achieves 95.0

The rapid advancement of Industry 5.0 and the concurrent growth of the Industrial Internet of Things (IIoT) present significant cybersecurity challenges necessitating advanced solutions. Digital Twin technology, which enables the creation of near-perfect digital replicas of physical systems, offers a promising approach to enhancing security and safety. This paper presents a literature review of the existing research to identify the challenges and future directions for integrating DT technology into IIoT from a security perspective. We aim to establish a comprehensive understanding of emerging features, including predictive analytics, real-time threat detection, and cybersecurity management. Additionally, this review highlights critical gaps, including complexity, model fidelity, real-time data processing, and scalability, which hinder the successful deployment of DT technology. Our study will assist researchers, cybersecurity practitioners, and policymakers in understanding the potential, limitations, and future advancements of this crucial area.

Federated learning (FL) is increasingly used in Industrial Internet of Things (IIoT) systems to support collaborative intelligence while preserving data privacy. However, adversarial clients and implicit trust assumptions can undermine reliability in safety-critical environments. Existing defense methods mainly rely on statistical or behavioral trust metrics, which often fail to detect semantically harmful yet statistically consistent updates, particularly under poisoning and backdoor attacks.To address these limitations, we propose ZTE-FL, a zero-trust federated learning framework that combines behavioral and semantic trust during aggregation. Behavioral trust captures temporal consistency and gradient alignment, while semantic trust evaluates functional correctness using clean and trigger-aware validation datasets. By enforcing continuous verification, ZTE-FL reduces reliance on implicit trust at both client and server levels.Across the evaluated datasets, attack scenarios, and model architectures, experimental results show that ZTE-FL achieved an average global accuracy of 97.92%, outperforming BC+FL+Trust (97.17%) and FedTrust (96.89%) across 18 attack-model combinations. ZTE-FL also achieved an average resilience score of 96.91% and reduced communication cost by 20.21% compared with TAIM. These results demonstrate that ZTE-FL improves robustness against poisoning and backdoor attacks while maintaining efficiency in latency-sensitive IIoT environments.

Nystagmus is a key indicator of vestibular disorders, including benign paroxysmal positional vertigo (BPPV). Accurate diagnosis of BPPV is essential, as it is treatable with specific bedside maneuvers that lead to rapid symptom resolution, thereby improving patient outcomes and reducing unnecessary treatments. In clinical practice, identification of positional nystagmus relies on eliciting and interpreting eye movements during provocative maneuvers, with or without video nystagmography (VNG). This process can be subjective and difficult to standardize when signals are subtle, noisy, or temporally variable. We present DSF-BPPVNet, a delay-aware neural architecture for BPPV classification from VNG traces. The model combines temporal convolution, delayed-state feedback, and residual refinement to support classification from temporally structured eye-movement signals. The model was evaluated on 3,111 VNG traces from 705 patients using 5-fold cross-validation and compared with established deep-learning baselines. In the patient-independent setting, DSF-BPPVNet achieved the strongest overall performance among the evaluated models, with an F1-score of 0.819 ± 0.020. Explainability analyses were also performed to characterize model attribution patterns and temporal weighting behavior.

The rapid growth of the Internet of Things (IoT) has significantly increased the complexity of device interactions, making IoT networks more vulnerable to sophisticated cyber threats. Effective intrusion detection is therefore crucial to ensuring the security and resilience of these systems. This paper presents federated learning with feature reduction (Fed-FeRe), a novel approach that enhances decentralized intrusion detection by integrating χ ^2 -based feature selection with a gated recurrent unit model. Fed-FeRe introduces an adaptive initialization of the performance threshold α and a data-driven estimation of key hyperparameters ( θ _0 , η ), enabling robust performance across diverse IoT conditions. By dynamically optimizing feature selection, the framework reduces computational overhead and communication costs, achieving approximately 2

Electroencephalogram (EEG) signal analysis plays a significant role in recognizing brain function and supporting the diagnosis of Epilepsy. Existing graph approaches are binary or use endpoint-only (slope/correlation) weights that do not capture the interior fluctuations and trend departures, which limit robustness for epilepsy detection. Many rely heavily on preprocessing and lack shift/scale robustness, which compromises generalization and reliability across subjects and recording conditions. The research aims to propose a novel SXTD-Weighted Visibility Graph framework with an information-rich edge weighting scheme to enhance interpretability and diagnostic accuracy in epilepsy EEG analysis. In addition, new EEG graph features such as Transitive Amplification Index (TAI), MedianWeightEps are developed that capturing the interior fluctuations and trend deviations, remaining shift-invariant and scale-equivariant, providing tunable noise–structure control, and requiring no additional assumptions. The proposed framework achieved 100

Provenance-based cybersecurity enables auditability and integrity in artificial intelligence systems but introduces substantial computational overhead, particularly for large language model (LLM) pipelines. Existing provenance compression approaches often sacrifice fidelity or lack demonstrable scalability. This paper presents CompProGraph, a two-stage lossless provenance compression framework that combines structural signature hashing with format-preserving JSON packing to reduce redundancy while preserving full W3C PROV semantics. Using the OpenML℄CC18 benchmark as a structural proxy for LLM-like provenance workflows, CompProGraph achieves consistent storage reductions of approximately 59% with linear-time compression and decompression. We further evaluate scalability on extended provenance corpora derived from repeated execution templates, ranging from megabytes to multi-gigabyte scale, demonstrating linear runtime and stable memory growth as corpus size increases by orders of magnitude. Although CC18 does not capture all semantic and temporal properties of real LLM pipelines, these results provide empirical evidence that exploiting structural similarity enables efficient, lossless provenance compression at scale. CompProGraph thus represents a practical step toward scalable and audit-ready provenance systems for trustworthy and compliant AI deployments.

Anomaly detection is crucial in high-performance computing (HPC) systems for maintaining effective, efficient, and secure operations. This survey focuses on the current status of the application of machine learning and deep learning in HPC systems for detecting various types of anomalies, including performance anomalies, operational anomalies, and security anomalies. The study takes a thorough look at the current approaches using diversified machine learning and deep learning techniques, the significance and challenges that anomaly detection in HPC systems brings, as well as the factors that should be considered in determining the performance of the systems according to the research conducted. Additionally, it explores tools and frameworks created using these techniques, specifically tailored for HPC systems. Nevertheless, it also reveals the issues with existing models, and based on them, further research is suggested. Hence, the discoveries unveiled in this study will be helpful for researchers and professionals specializing in anomaly detection within HPC systems.

The rapid proliferation of Large Language Models (LLMs) across industries such as healthcare, finance, and legal services has revolutionized modern applications. However, their increasing adoption exposes critical vulnerabilities, particularly through adversarial prompt attacks that compromise LLM security. These prompt-based attacks exploit weaknesses in LLMs to manipulate outputs, leading to breaches of confidentiality, corruption of integrity, and disruption of availability. Despite their significance, existing research lacks a comprehensive framework to systematically understand and mitigate these threats. This paper addresses this gap by introducing a taxonomy of prompt attacks based on the Confidentiality, Integrity, and Availability (CIA) triad, an important cornerstone of cybersecurity. This structured taxonomy lays the foundation for a unique framework of prompt security engineering, which is essential for identifying risks, understanding their mechanisms, and devising targeted security protocols. By bridging this critical knowledge gap, the present study provides actionable insights that can enhance the resilience of LLM to ensure their secure deployment in high-stakes and real-world environments.

Ensuring the security and integrity of Federated Learning (FL) models against adversarial attacks is critical. Among these threats, targeted data poisoning attacks, particularly label flipping, pose a significant challenge by undermining model accuracy and reliability. This paper investigates targeted data poisoning attacks in FL systems, where a small fraction of malicious participants corrupt the global model through mislabeled data updates. Our findings demonstrate that even a minor presence of malicious participants can substantially decrease classification accuracy and recall, especially when attacks focus on specific classes. We also examine the longevity and timing of these attacks during early and late training rounds, highlighting the impact of malicious participant availability on attack effectiveness. To mitigate these threats, we propose a defense strategy that identifies malicious participants by analyzing parameter updates across vulnerable training rounds. Utilizing Principal Component Analysis (PCA) for dimensionality reduction and anomaly detection, our approach effectively isolates malicious updates. Extensive simulations on standard datasets validate the effectiveness of our algorithm in accurately identifying and excluding malicious participants, thereby enhancing the integrity of the FL model. These results offer a robust defense against sophisticated poisoning strategies, significantly improving FL security.

Deepfake technology enables the creation of highly realistic but fabricated videos, raising significant concerns about the authenticity and reliability of digital media. As deepfakes become increasingly sophisticated, effective detection methods have become crucial. One promising approach involves Remote PhotoPlethysmography (rPPG), a non-contact method that measures subtle changes in skin color to estimate heart rate. rPPG is a powerful tool for detection as deepfake videos often fail to replicate the physiological signals inherent in real human subjects. This survey aims to provide a comprehensive understanding on various rPPG-based techniques for identifying deepfakes and analyzing their effectiveness, challenges, and future potential. To improve the detection accuracy, methods that enhance rPPG signals in combination with machine learning models are examined. Real-world applications, including the detection of fake celebrity videos that are employed in propaganda, illustrate the applicability of these techniques. Through a comprehensive review of existing rPPG-based deepfake detection techniques, this survey aims to inform future research and development endeavors in the area, thereby contributing to broader efforts to secure digital content from tampering and exploitation.

Intrusion Detection Systems (IDS) are essential for securing computer networks by identifying and mitigating potential threats. However, traditional IDS systems face challenges related to scalability, privacy, and computational demands as network data complexity increases. Federated Learning (FL) has emerged as a promising solution, enabling collaborative model training on decentralized data sources while preserving data privacy. Each participant retains local data repositories, ensuring data sovereignty and precluding data sharing. Leveraging the FL framework, participants locally train machine learning models on their respective datasets, subsequently transmitting model updates to a central server for aggregation. The central server then disseminates the aggregated model updates to individual participants, collectively striving to bolster intrusion detection capabilities. This paper presents a comprehensive survey of FL applications in IDS, covering core concepts, architectural approaches, and aggregation strategies. We evaluate the strengths and limitations of various FL methodologies for IDS, addressing privacy and security concerns and exploring privacy-preserving techniques and security protocols. Our examination of aggregation strategies within the FL framework for IDS aims to highlight their effectiveness, limitations, and potential enhancements.

The rapid expansion of IoT devices has introduced significant cybersecurity risks, as attackers increasingly exploit these networks’ vulnerabilities. To counter this threat, this paper presents the Privacy-Enhanced IoT Defence System (PEIoT-DS), a novel solution that emphasises data privacy while delivering high-performance intrusion detection. PEIoT-DS use federated learning to create a comprehensive intrusion detection model without necessitating the transmission of raw data to a central server. IoT devices only contribute model updates, which are then combined to improve the global model. While allowing devices to benefit from the network’s collective insights, this decentralised learning methodology safeguards data privacy. Using a real-world IoT dataset and two popular federated learning algorithms—Federated Average and Federated Average with Momentum—the study assesses the effectiveness of PEIoT-DS. The findings show that, in comparison to Federated Average, Federated Average with Momentum produces faster convergence and better intrusion detection accuracy. Our PEIoT-DS approach offers a reliable intrusion detection system for IoT networks while maintaining privacy.

The escalating sophistication of cyber threats poses significant risks to individuals, organizations, and nations. Cybercrime, encompassing activities like hacking and data breaches, has severe economic and societal consequences. In today’s interconnected world, robust cybersecurity measures are paramount to mitigate these risks and protect sensitive information. However, traditional security solutions struggle to keep pace with the evolving threat landscape. Artificial Intelligence (AI) offers a powerful arsenal of techniques to address these challenges. This paper explores the application of AI methods, including Machine Learning (ML), Deep Learning (DL), Natural Language Processing (NLP), Explainable AI (XAI), and Generative AI, in solving various cybersecurity problems. This paper presents a comprehensive analysis of AI techniques for enhancing cybersecurity. Key contributions include: 1) comparative study of ML and DL methods: Evaluating their accuracy, applicability, and suitability for various cybersecurity challenges; 2) investigation into XAI approaches: Enhancing the transparency and interpretability of AI-powered security solutions, particularly in anomaly detection; 3) exploration of emerging trends in Generative AI (Gen-AI) and NLP: Examining their potential to simulate and mitigate cyber threats through advanced techniques like threat intelligence generation and attack simulations; 4) application of GenAI in cybersecurity and real-world products of GenAI for cyber security. This research aims to advance the state-of-the-art in AI-driven cybersecurity by providing insights into effective and reliable solutions for mitigating cyber risks and improving the overall security posture.

This paper presents a comprehensive, cross-sector analysis of large-scale ICT failures to address the persistent gap in understanding how systemic digital breakdowns occur and propagate across platforms and industries. Through a comparative study of seven major global outages (2019–2024) — selected based on scale, technical transparency, and platform diversity — we identify recurring vulnerabilities in automation governance, configuration management, centralized infrastructure, and incident response. Using a custom analytical framework grounded in socio-technical and resilience engineering theory, the paper maps failure propagation patterns and derives a taxonomy of technical and organizational failure modes.We empirically validate a suite of resilience strategies — including rollback automation, configuration-as-code, SOAR-enabled response orchestration, and chaos engineering — and demonstrate how they address failure propagation pathways observed in real-world incidents. A conceptual model for decentralized system upgrade planning is introduced, incorporating microservice segmentation, dependency mapping, and AI-assisted fault containment. The paper culminates in a forward-looking digital resilience roadmap that integrates predictive analytics, secure software supply chains, and adaptive human–machine collaboration. Core contributions include: (1) a cross-case classification of failure archetypes, (2) evidence-based design patterns for resilience, and (3) actionable frameworks for infrastructure operators and researchers working towards next-generation ICT robustness.

The Industrial Internet of Things (IIoT) is transforming industrial operations through connected devices and real-time automation but also introduces significant cybersecurity risks. Cyber threat intelligence (CTI) is critical for detecting and mitigating such threats, yet traditional centralized CTI approaches face limitations in latency, scalability, and data privacy. Federated learning (FL) offers a privacy-preserving alternative by enabling decentralized model training without sharing raw data. This survey explores how FL can enhance CTI in IIoT environments. It reviews FL architectures, orchestration strategies, and aggregation methods, and maps their applications to domains such as intrusion detection, malware analysis, botnet mitigation, anomaly detection, and trust management. Among its contributions is an empirical synthesis comparing FL aggregation strategies—including FedAvg, FedProx, Krum, ClippedAvg, and Multi-Krum—across accuracy, robustness, and efficiency under IIoT constraints. The paper also presents a taxonomy of FL-based CTI approaches and outlines future research directions to support the development of secure, scalable, and decentralized threat intelligence systems for industrial ecosystems.