The increasing prevalence of adversarial evasion techniques poses a significant challenge to the reliability of intrusion detection systems (IDSs) in Internet-of-Things (IoT) environments. Minor manipulations of traffic telemetry can alter temporal behaviour or frequency-level patterns, leading to degraded detection performance and potential security failures. This study presents the Multi-Frequency Temporal Spatio-Transformer (MFTST), a computational framework designed to improve adversarial robustness in IoT intrusion detection. MFTST combines temporal Transformer-based encoding with multi-frequency channel attention to capture both sequential traffic behaviour and frequency-level variations in learned traffic representations. The framework is evaluated using a Composite attack that emulates the detection-side effects of two representative defense-evasion behaviours: obfuscation (T1027) and indicator tampering (T1070), as commonly described in the MITRE ATT&CK knowledge base. The Composite attack jointly perturbs temporal attention behaviour and frequency-domain traffic representations to assess the resilience of MFTST against timing-based masking, attention disruption, and frequency-domain perturbations. To improve robustness, the model is trained using a composite adversarial optimization strategy that jointly considers classification performance, temporal-attention stability, and frequency-domain perturbation control. Experimental evaluations on MQTTset and X-IIoTID demonstrate that MFTST improves detection performance under both clean and adversarial conditions. The results show that adversarial training improves multiclass accuracy from 0.8314 to 0.8558 on MQTTset and from 0.8758 to 0.9192 on X-IIoTID. Under stronger adversarial evaluation, MFTST also maintains higher robustness than conventional machine-learning, deep-learning, and robustness-oriented baselines. These findings indicate that temporally conditioned frequency attention can improve both detection reliability and interpretability in adversarial IoT security monitoring.

Context: Software vulnerability detection (SVD) is increasingly challenged by the scale and complexity of modern software systems. Although deep learning–based approaches demonstrate strong detection performance, their adoption in security-critical settings is limited by insufficient interpretability, adversarial resilience, and systematic Responsible AI integration. Objective: This paper aims to design and evaluate a Responsible AI–driven framework for SVD that operationalizes fairness, interpretability, security, reliability, and transparency without compromising detection effectiveness. Method: We propose a model-agnostic vulnerability detection framework that incorporates fairness-aware data preprocessing, multi-model evaluation, and structured verification mechanisms. Interpretability is achieved through multi-view explanations combining global and local SHAP, LIME, and attention-based attribution. Explanation consistency is examined using attention-guided token occlusion, while security is evaluated via multiple white-box adversarial attacks on correctly classified test samples. The framework is validated on three public datasets—CWE-119, CWE-399, and DiverseVul—using deep learning and pre-trained code models. Results: Experimental results demonstrate competitive detection performance across datasets while providing structured explanation and adversarial evaluation evidence. Interpretability analyses reveal dataset-specific vulnerability cues aligned with domain knowledge, and perturbation studies show consistent confidence shifts under controlled token masking. Adversarial experiments further illustrate stable performance trends under attack conditions. Conclusion: The findings indicate that Responsible AI principles can be systematically operationalized within deep learning–based SVD pipelines. By integrating fairness, interpretability, security evaluation, reliability assessment, and transparency mechanisms, the proposed framework supports trustworthy and security-aware deployment of AI-driven vulnerability detection systems.

The rapid expansion of IoT devices in smart supply chains has intensified system complexity and widened the attack surface, exposing resource-constrained edge networks to threats such as DoS and data poisoning. Addressing these challenges requires intrusion detection methods that are both computationally efficient and structurally robust. To this end, we propose a novel Spectral Relational Representation Distillation (SPRRD) framework that compresses relational knowledge by projecting embeddings onto a low-rank spectral subspace using power iteration and QR-based orthonormalization. This preserves relational geometry while substantially reducing memory and computation. To further enhance robustness, we introduce Weighted Spectral Relational Representation Distillation (WSPRRD), which employs a trainable spectral weighting vector to prioritize informative directions and adapt to noise and evolving attack behaviors. Using a BiGRU teacher–student design, we evaluate both methods on large-scale IoT intrusion datasets, i.e., CICDDoS2019 and CICIoT2023, with standard metrics and deeper spectral analyses, including cosine similarity, energy retention, and effective rank. Results show that SPRRD accelerates training with negligible accuracy loss, while WSPRRD strengthens relational alignment and generalization under noisy and dynamic conditions. These findings demonstrate that spectral-aware relational distillation offers a scalable, resilient, and computation-efficient solution for modern IoT-based smart supply chain security.

The evolution of Industry 5.0 introduces complex, temporally irregular network behaviors, challenging intrusion detection in heterogeneous industrial environments. Existing Intrusion Detection Systems (IDSs) often struggle with asynchronous traffic and lack interpretability. To overcome these limitations, we propose AegisNet, a novel sequence-based IDS that reformulates intrusion detection as a sequence classification task on fixed-length windows ofTconsecutive observations (sequence length), augmented with explicit time-delta features to capture irregular traffic intervals. AegisNet introduces a novel Temporal-Aware and Self-Adaptive Multi-Head Attention (TA-SAMHA) mechanism. It introduces learnable temporal bias projections to directly encode irregular time intervals into the attention process, enabling the system to capture fine-grained temporal dependencies and evolving behaviors. It also employs a Dual-Gate Residual Encoder (DGRE) for bidirectional sequence modeling with dual gating, enabling enriched contextual representations and stable learning across deep temporal layers. Additionally, we introduce the IDS-specific adaptation of Temporal SHAP (T-SHAP), enabling time-resolved feature attributions that expose how influential factors shift during sequential intrusion detection, offering a new dimension of temporally aware explainability for evolving attack patterns. Experimental results demonstrate that AegisNet achieves a detection accuracy of 99.63% on the CICIoV2024 dataset and 99.13% on the CICIoT2023 dataset, demonstrating its effectiveness for securing Industrial environments.

Early detection and proper treatment of epilepsy is essential and meaningful to those who suffer from this disease. The adoption of deep learning (DL) techniques for automated epileptic seizure detection using electroencephalography (EEG) signals has shown great potential in making the most appropriate and fast medical decisions. However, DL algorithms have high computational complexity and suffer low accuracy with imbalanced medical data in multi seizure-classification task. Motivated from the aforementioned challenges, we present a simple and effective hybrid DL approach for epileptic seizure detection in EEG signals. Specifically, first we use a K-means Synthetic minority oversampling technique (SMOTE) to balance the sampling data. Second, we integrate a 1D convolutional neural network (CNN) with a Bidirectional Long Short-Term Memory (BiLSTM) network based on Truncated Backpropagation Through Time (TBPTT) to efficiently extract spatial and temporal sequence information while reducing computational complexity. Finally, the proposed DL architecture uses softmax and sigmoid classifiers at the classification layer to perform multi and binary seizure-classification tasks. In addition, the 10-fold cross-validation technique is performed to show the significance of the proposed DL approach. Experimental results using the publicly available UCI epileptic seizure recognition data set shows better performance in terms of precision, sensitivity, specificity, and F1-score over some baseline DL algorithms and recent state-of-the-art techniques.

Deterministic and low-latency communication enabled by Time-Sensitive Networking (TSN) is essential for modern large-scale segmented industrial networks, which are typically divided into multiple operational domains. However, securing such multi-domain, functionally segmented TSN environments remains challenging due to resource-constrained edge devices, strict real-time requirements, and uneven availability of labeled attack data across isolated domains. To address these challenges, we propose a Selective Federated Intrusion Detection Framework with Domain Adaptation (SAFID) tailored for large-scale segmented industrial networks. The framework introduces a cluster-aware domain classification strategy that identifies data-rich (active) and data-scarce (passive) domains based on Quality-of-Service (QoS) flow density. Active domains collaboratively train TDNet, a compact neural network architecture optimized for TSN traffic, through federated learning. TDNet employs low-rank factorized dense layers to reduce computational overhead and parameter count while preserving classification accuracy. To further minimize latency and memory consumption, a post-training quantization step compresses both weights and activations. Passive domains receive the globally aggregated model and locally fine-tune it via a lightweight domain-specific adaptation mechanism to capture localized threats. A post-training quantization step further compresses TDNet for deployment in latency-critical industrial environments. Extensive evaluation of SAFID demonstrates detection accuracies of 99.75% on CICIDS2017 and 95.75% on Edge-IIoTset. The framework also achieves a 6.5× improvement in inference throughput after quantization and exhibits robust generalization across diverse domain-specific attack patterns. Unlike prior methods treating federated intrusion detection uniformly, our framework enables selective participation and targeted adaptation, making it highly suitable for real-time, resource-constrained multi-domain TSN networks.

The rapid proliferation of Internet of Things (IoT) devices has transformed modern infrastructures, yet their inherently distributed and dynamic nature poses significant challenges for cybersecurity. Traditional intrusion detection systems (IDS) often rely on static models or linear temporal analysis, which makes them insufficient to identify and respond to evolving threats that manifest across both spatial and temporal dimensions. Furthermore, existing attention-based mechanisms tend to treat spatial and temporal dependencies separately and apply fixed attention weights, limiting their adaptability in complex IoT environments. To address these limitations, we propose, a novel IDS framework incorporating a Spatio-Temporal Conditioning with Dynamic Multi-Head Attention (STC-MHA-DW) mechanism. This module captures contextual interdependencies across both spatial features and temporal windows by leveraging head-wise adaptive projections, enabling the model to dynamically reweight attention based on surrounding threat context. The temporal encoder is built using a dual-stage gated mechanism that processes both short-term fluctuations and long-term dependencies, while the attention layer refines representations through localized spatio-temporal salience. We also introduce a scalable, cloud-native deployment architecture using microservices and containerization to ensure efficient performance under dynamic network loads. Experimental evaluations show that the proposed ids achieves the highest detection accuracy of 99.84%, precision of 99.47%, recall of 98.83%, and f1-score of 99.14% with a very low false alarm rate, outperforming existing models.

The increasing adoption of Unmanned Aerial Ve-hicles (UAV s) in various critical applications necessitates robust security measures to protect these systems from cyber threats. In response, this research introduces an innovative Intrusion Detection System (IDS) specifically tailored for UAV s. The proposed IDS leverages Hierarchical Attention-based Long Short-Term Memory (H-LSTM) networks to effectively model the intricate temporal dependencies in UAV data. This architecture allows for comprehensive surveillance of UAV behavior, capturing both short-term anomalies and long-term deviations from expected patterns. The hierarchical attention mechanism enables the system to focus on salient features within the data, enhancing detection accuracy and robustness. To address the critical need for interpretable AI in cybersecurity, we incorporate Shapley Ad-ditive Explanations (SHAP) into our IDS. SHAP values provide a coherent and intuitive explanation of the IDS's decisions by emphasizing the specific features and their contributions to the intrusion detection process. The performance of the proposed system is rigorously evaluated using the N-BaIoT dataset. Our experiments demonstrate that the H-LSTM-based IDS outper-forms traditional methods, achieving a higher detection rate while minimizing false positives. Moreover, the incorporation of SHAP explanations facilitates rapid incident analysis, allowing security professionals to discern between genuine threats and benign anomalies effectively.

The exponential growth of Internet of Things (IoT) devices poses distinctive challenges to safeguarding the security and privacy of interconnected systems. As the frequency of cyberattacks continues to escalate, the development of an effective and scalable Intrusion Detection System (IDS) based on Federated Learning (FL) for IoT becomes increasingly complex. Existing methodologies struggle to balance spatial and temporal feature extraction, particularly when confronted with dynamic and evolving cyber threats. Additionally, the lack of diversity in datasets employed for FL-based IDS evaluations further hinders progress. Furthermore, a notable tradeoff between performance and scalability emerges, particularly as the number of edge devices in communication grows. Motivated by the aforementioned challenges, this article presents a horizontal FL model that combines Convolutional Neural Networks (CNN) and Bidirectional Long-Term Short Memory (BiLSTM) for effective intrusion detection. This hybrid approach aims to address the limitations of existing methods and enhance the effectiveness of intrusion detection in the context of FL for IoT. Specifically, CNN plays a pivotal role in spatial feature extraction, allowing the model to identify and comprehend local patterns indicative of potential intrusions, and the BiLSTM component contributes by capturing temporal dependencies and learning sequential patterns within the data. The proposed IDS adheres to a zero-trust model by keeping the data on local edge devices, sharing only the learned weights with the centralized FL server. In turn, the FL server aggregates updates from diverse sources to optimize the accuracy of the global learning model. The experimental results using CICIDS2017 and Edge-IIoTset prove the effectiveness of the proposed approach over centralized and federated deep learning-based IDS.

The Internet of Things (IoT) has revolutionized various sectors by enabling seamless device interaction. However, the proliferation of IoT devices has also raised significant security and privacy concerns. Traditional security measures often fail to address these concerns due to the unique characteristics of IoT networks, such as heterogeneity, scalability, and resource constraints. This survey paper adopts a thematic exploration approach for a comprehensive analysis to investigate the convergence of quantum computing, federated learning, and 6G wireless networks. This novel intersection is explored to significantly improve security and privacy within the IoT ecosystem. To enable several secure, intelligent IoT applications, quantum computing, with its superior computational capabilities, can strengthen encryption algorithms, making IoT data more secure. Federated learning, a decentralized machine learning approach, allows IoT devices to learn a shared model while keeping all the training data on the original device, thereby enhancing privacy. This synergy becomes even more crucial when integrated with the high-speed, low-latency capabilities of 6G networks, which can facilitate real-time, secure data processing and communication among many IoT devices. Second, we discuss the latest developments, offering an up-to-date overview of advanced solutions, available datasets, and key performance metrics and summarizing the vital insights, challenges, and trends in securing IoT systems. Third, we design a conceptual framework for integrating quantum computing in federated learning, adapted for 6G networks. Finally, we highlight the future advancements in quantum technologies and 6G networks and summarize the implications for IoT security, paving the way for researchers and practitioners in the field of IoT security.

The Industrial Internet of Things (IIoT) is rapidly evolving, and with this evolution, cyber threats have become a significant issue. IIoT networks, despite improving service quality, are uniquely vulnerable to security threats due to their inherent connectivity and the use of low-power devices. Traditional Deep Learning-based IDS, while accurate, suffer from a “black box” issue that hides the reasoning behind their decisions, leading to a decrease in user trust. To address this, our research presents an Explainable and intelligent mechanism for data-efficient intrusion detection in IIoT. Our proposed IDS enhances data efficiency by employing a Bidirectional Long-Short Term Memory (BiLSTM) model with a self-adaptive attention mechanism. The selfadaptive attention mechanism is a novel feature of our IDS framework, designed specifically for IIoT environments. This mechanism dynamically adjusts its focus to prioritize critical elements within a dataset, allocating more computational resources to data segments likely to contain patterns or anomalies indicative of security threats. When integrated with BiLSTM, which excels at capturing temporal dependencies, the mechanism enhances the IDSs ability to learn efficiently from limited datasets. This focus on significant data features and temporal patterns reduces the need for extensive training datasets, making it particularly effective in IIoT settings where data may be sparse yet complex. In addition, we enhance the proposed IDSs transparency by incorporating the SHapley Additive exPlanations mechanism from Explainable AI, thereby boosting the IDSs trustworthiness and interpretability. Our system exhibits outstanding performance on benchmark datasets such as CICIDS2017 and X-IIoTID, attaining accuracies of 99.92% and 96.54%, respectively.

The deployment of Internet of Things (IoT) systems in Smart Agriculture (SA) operates in extreme environments including wind, snowfall, flooding, landscape, and so on for collecting and processing real-time data. The increased connectivity and broad adoption of IoT devices with low-power communications on farmland support farmers in making data-driven decisions using various Artificial Intelligence (AI) techniques. Furthermore, in such an environment, edge computing is also utilized to provide computationally intensive, latency-sensitive, and bandwidth-demanding services at the edge of the network. However, protecting edge-to-Things in the extreme environment of SA is challenging, due to the volume of data, and also attackers exploit network gateways to perform Distributed Denial of Service (DDoS) attacks. Motivated by the aforementioned challenges, we develop a novel deep learning-based Intrusion Detection System (IDS) for edge-envisioned SA in extreme environments. Specifically, a hybrid approach is developed by combining bidirectional gated recurrent unit, long-short term memory with softmax classifier to detect attacks at the edge of the network. To allow faster learning, the proposed IDS employs the Truncated Backpropagation through Time (TBPTT) approach to handle lengthy sequences of network data. Furthermore, we suggest an attack scenario with deployment architecture for the proposed IDS in the extreme environment of SA. Extensive experiments using three publicly available datasets namely, CIC-IDS2018, ToN-IoT, and Edge-IIoTset prove the effectiveness of the proposed IDS over some traditional and contemporary state-of-the-art techniques.

Summary Artificial Intelligence (AI) based cyber threat detection tools are widely used to process and analyze a large amount of data for improved intrusion detection performance. However, these models are often considered as black box by the cybersecurity experts due to their inability to comprehend or interpret the reasoning behind the decisions. Moreover, AI‐based threat hunting is data‐driven and is usually modeled using the data provided by multiple cloud vendors. This is another critical challenge, as a malicious cloud can provide false information (i.e., insider attacks) and can degrade the threat‐hunting capability. In this paper, we present a blockchain‐enabled eXplainable AI (XAI) for enhancing the decision‐making capability of cyber threat detection in the context of Smart Healthcare Systems. Specifically, first, we use blockchain to validate and store data between multiple cloud vendors by implementing a Clique Proof‐of‐Authority (C‐PoA) consensus. Second, a novel deep learning‐based threat‐hunting model is built by combining Parallel Stacked Long Short Term Memory (PSLSTM) networks with a multi‐head attention mechanism for improved attack detection. The extensive experiment confirms its potential to be used as an enhanced decision support system by cybersecurity analysts.

The efficient patient-independent and interpretable framework for electroencephalogram (EEG) epileptic seizure detection (ESD) has informative challenges due to the complex pattern of EEG nature. Automated detection of ES is crucial, while Explainable Artificial Intelligence (XAI) is urgently needed to justify the model detection of epileptic seizures in clinical applications. Therefore, this study implements an XAI-based computer-aided ES detection system (XAI-CAESDs), comprising three major modules, including of feature engineering module, a seizure detection module, and an explainable decision-making process module in a smart healthcare system. To ensure the privacy and security of biomedical EEG data, the blockchain is employed. Initially, the Butterworth filter eliminates various artifacts, and the Dual-Tree Complex Wavelet Transform (DTCWT) decomposes EEG signals, extracting real and imaginary eigenvalue features using frequency domain (FD), time domain (TD) linear feature, and Fractal Dimension (FD) of non-linear features. The best features are selected by using Correlation Coefficients (CC) and Distance Correlation (DC). The selected features are fed into the Stacking Ensemble Classifiers (SEC) for EEG ES detection. Further, the Shapley Additive Explanations (SHAP) method of XAI is implemented to facilitate the interpretation of predictions made by the proposed approach, enabling medical experts to make accurate and understandable decisions. The proposed Stacking Ensemble Classifiers (SEC) in XAI-CAESDs have demonstrated 2% best average accuracy, recall, specificity, and F1-score using the University of California, Irvine, Bonn University, and Boston Children's Hospital-MIT EEG data sets. The proposed framework enhances decision-making and the diagnosis process using biomedical EEG signals and ensures data security in smart healthcare systems.

The recent innovations in the network communication domain have entirely revolutionized the conventional industrial sector by introducing a new era of automatic communication. The Industrial Internet of Things (IIoT) is acknowledged as an exclusive space where Internet of Things (IoT) seems to divulge expressive manifestations. However, the expanded connectivity, more openness, and the widespread use of low-power communication devices of IIoT makes them vulnerable to malicious attacks, and criminal activities. Moreover, the heterogeneous and prevalent nature of the IIoT devices makes it very difficult to come up with a centralized threat detection mechanism, thus, its security remains a major concern. Motivated by this, we propose a fog-empowered augmented intelligence (IA)-based defensive mechanism to ensure secure communication in IoT-enabled smart industries. The designed framework incorporates the aggregated potentials of two highly acclaimed deep learning (DL) classifiers: 1) gated recurrent unit (GRU) and 2) bidirectional long short-term memory (BiLSTM), where the DL-based scheme detects anomalies in such industrial networks and the FOG-based scenario promotes the routing flexibility and interoperability of the heterogeneous devices of the IoT-enabled smart industrial network. The validity of the proposed mechanism is analytically investigated by conducting a comparison with the benchmark threat detection techniques. The proposed solution is also generically examined parallel to some state-of-the-art approaches. The Cu-GRU-BiLSTM framework achieved up to 99.91% accuracy with a low-false alarm rate and outperforms the baseline and recent threat detection approaches. The simulation and comparison results validate the effectiveness of the proposed mechanism and advocate it as a phenomenal choice to ensure efficacious and secure communication in IoT-based smart industries.