Management of Sensor-Temporal Threats AIML-Based Explainable IDS for Cyber-Physical Systems
Аннотация
Timing sensor attacks have become a critical threat to cyber-physical systems (CPS), exploiting synchronization, correlation, and multi-sensor manipulation in ways that traditional IDS solutions cannot detect. Sensor-temporal attacks have emerged as a critical threat to Cyber-Physical Systems (CPS), exploiting timing, correlation, and multi-sensor manipulation in ways traditional IDS solutions cannot detect. To describe the various CPS modalities and their structural–temporal connections, this study aimed to develop a unified, explainable AIML-based intrusion detection system. We examined baseline, hybrid, and advanced architectures in an experimental evaluation using multimodal public datasets, such as SWaT, WADI, ToN-IoT, CIC-IDS2017, and N-BaIoT. — The proposed Sensor-Temporal Graph Transformer (ST-GT) combines fidelity-focused explainability modules, contrastive multimodal pre-training, and graph-aware attention. Fidelity-driven explainability modules, contrastive multimodal pretraining, and graph-aware attention. Outperforming the Hybrid and LSTM benchmarks, the ST-GT achieved the best performance on all parameters, including F1 = 0. Baselines, ST-GT had the best performance across all parameters, including F1 = 0. 93, ROC-AUC = 0. 97 and PR-AUC = 0. 97, and PR-AUC = 0. 95. The explainability Results further improved with counterfactual plausibility = 0. for explainability further improved with counterfactual plausibility = 0. 83, fidelity = 0. 81 et AUC d'insertion = 0. 81, and insertion AUC = 0. 82, indicating robust and interpretable decision-making behavior. Decision behaviour. These results demonstrate that integrating graphical temporal reasoning with explainable AI significantly improves intrusion detection in cyber-physical systems (CPS), providing a highly accurate and reliable solution suitable for real-time industrial implementation. The integration of graph-temporal reasoning with explainable AI substantially improves intrusion detection in cyber-physical systems (CPS), providing a highly accurate and reliable solution appropriate for real-time industrial implementation