Enhancing Cybersecurity in Cyber-Physical Systems: an Explainable AI Approach for Intrusion Detection

Cyber-Pyshical Systems (CPS) are increasingly vulnerable to worst-case cyberattacks, where adversaries exploit both cyber and physical components to maximize disruption while avoiding detection. Traditional Intrusion Detection Systems (IDS) struggled against these adaptive threats due to their resilience on static rules and black-box models. This paper proposes an Explainable Artificial Intelligence-Based Intrusion Detection System (XAI-IDS) tailored for CPS security. By leveraging SHapley Additive Explanations (SHAP) and Local Interpretable Model-Agnostic Explanations (LIME), our approach enhances threat visibility, enabling security practitioners to understand, trust, and refine attack mitigation strategies. The system dynamically detects cyber-physical anomalies, prioritizing high-risk attack paths, that reduce system resilience by 19% in worst-case scenarios. Experimental results show that Random Forest achieves 98.2% accuracy, Decision Tree 95.7%, and Logistic Regression 91.4%, with SHAP analysis identifying network traffic rate, sensor anomalies, and unauthorized access attempts as the most critical threat indicators. By reducing detection latency and improving interpretability, the proposed XAI-IDS ensures proactive real-time security industrial control systems against sophisticated adversarial threats.