SmartCPS-ADAPT: An Intrusion Detection System for Cyber-Physical Systems

Abstract

The rapid growth of cyber-physical systems (CPS) and internet of things Edge-Cloud environments has amplified concerns regarding cyber threats, making intrusion detection systems (IDS) a vital component of network resilience. However, recent studies reveal a research gap in existing IDS frameworks: many fail to adapt to evolving attacks, exhibit imbalanced detection performance, or incur high computational costs. To address this challenge, this work aims to develop a robust deep learning-based classification model that accurately detects cyberattacks in CPS. Hence, this work proposed SmartCPS-ADAPT, which leverages Convolutional Neural Network-Bi-Directional Long Short-Term Memory (CNN–BiLSTM) for spatial-temporal feature extraction, an adaptive extreme gradient boosting (XGBoost) classifier for handling evolving data distributions, and a drift detection mechanism for continuous learning. Experimental evaluation on the CICIoT2023 dataset demonstrates superior performance, achieving 100% accuracy in binary classification and 99.85% accuracy in multi-class classification. The findings confirm that SmartCPS-ADAPT significantly outperforms existing approaches, ensuring reliable detection of diverse cyber attacks. In conclusion, the SmartCPS-ADAPT establishes a highly effective IDS model that addresses adaptability, robustness, and accuracy in CPS security.