CPS-IoT-PPDNN: A new explainable privacy preserving DNN for resilient anomaly detection in Cyber-Physical Systems-enabled IoT networks

The integration of Cyber-Physical Systems (CPS) within the Internet of Things (IoT) ecosystem has transformed various sectors, enabling intelligent, interconnected environments that blend computational and physical processes. However, the security and privacy vulnerabilities within CPS-IoT networks remain critical, as anomalies can lead to severe, system-wide consequences. To address these challenges, this research introduces a novel, explainable, privacy-preserving Deep Neural Network (DNN) framework for anomaly detection in CPS-enabled IoT networks. While deep learning models are widely used in Intrusion Detection Systems (IDSs) for their capability to analyze vast data sources, their high false-positive rates and lack of interpretability present limitations. Our framework, therefore, employs a deep SHpley Additive exPlanations (SHAP) technique to clarify the DNN's decision-making process, aiding users and cybersecurity experts in validating and reinforcing the system's resilience. This approach was tested on two state-of-the-art datasets—Edge-IIoTset and X-IIoTID—demonstrating outstanding results. For binary classification, both datasets achieved 100 % accuracy, precision, recall, and F1-score, while multi-class scenarios reached nearly perfect metrics, with Edge-IIoTset achieving 99.98 % accuracy and X-IIoTID achieving 99.99 %. Additionally, our model showed significantly faster training times without compromising testing efficiency. The results confirm that this proposed explainable DNN framework offers robust, real-time, and privacy-preserving intrusion detection, enhancing CPS-IoT networks' defenses against advanced cyber threats.