On the Performance of GRASP-Based Feature Selection for CPS Intrusion Detection

Cyber-Physical Systems (CPS) are the basis for the world’s critical infrastructure and, thus, have the potential to significantly impact human lives in the near future. In recent years, there has been an increasing demand for connectivity in CPS, which has brought to attention the issue of cybersecurity. Aside from traditional information systems threats, CPS face new challenges due to the heterogeneity of devices and protocols, as well as its strong reliability requirements. In this work, we provide a brief overview of the CPS architecture and applications and describe the security challenges in the three CPS layers of perception, transmission, and application. Besides, we discuss how feature selection (FS) may improve intrusion detection performance. In particular, we evaluate how metaheuristic approaches can improve classification performance in CPS perception, transmission, and application layers. Our results reveal that (i) Greedy Randomized Adaptive Search Procedure (GRASP) outperforms traditional filter-based methods, and (ii) using the proposed enhanced approaches in GRASP construction and local search phases can enhance the average F1-Score of five classifier algorithms.