NGMO: A Novel Geometric Mean Optimizer for Intrusion Detection in Industrial Cyber-Physical Systems

Abstract

Industrial cyber-physical systems (CPS) are experiencing various malicious attacks and encountering increasing security challenges. Although machine learning-based intrusion detection systems can help users quickly detect attacks in industrial CPS, feature redundancy and the tuning of model hyperparameters hinder further detection performance. In this study, a Novel Geometric Mean Optimizer (NGMO) is designed to filter redundant industrial features while optimizing the hyperparameters of model. The proposed NGMO incorporates good point sets and dynamic opposition learning strategies during the population initialization and generation hopping phases to enhance the search capabilities of algorithm. Furthermore, the NGMO is combined with three gradient boosting decision tree models for intrusion detection in industrial CPS. Finally, four datasets from industrial scenarios and a real-world case are used to evaluate the effectiveness of NGMO. The experimental results show that NGMO can reduce time consumption while improving model detection accuracy. Therefore, the proposed NGMO can effectively enhance the security of industrial CPS.