MetaCAN: An Optimized Adaptive Hybrid Metaheuristic-based Intrusion Detection System for CAN Bus Security

The Controller Area Network (CAN) bus is a message-based protocol widely used in modern vehicles to facilitate communication between various Electronic Control Units (ECUs). However, its simplistic design lacks fundamental security measures, making it highly susceptible to cyberattacks. These vulnerabilities pose significant risks to vehicle safety, highlighting the critical need for implementation of effective intrusion detection systems (IDS). Therefore, in this paper, a machine learning based IDS optimized through an adaptive hybrid metaheuristic approach, named MetaCAN, is proposed to secure the CAN bus. MetaCAN leverages the complementary strengths of particle swarm optimization (PSO) for fast convergence and cuckoo search (CS) for robust global search to ensure effective hyperparameter tuning and model optimization. MetaCAN is evaluated using three real-world datasets including Survival Analysis, Car Hacking: Attack \& Defense Challenge 2020, and OTIDS. Unlike traditional binary detection systems, MetaCAN offers multi-class attack detection by identifying five distinct attack types including Denial of Service (DoS), fuzzy, masquerade, malfunction, and replay attacks. Moreover, the detection accuracy of the system is enhanced through a feature engineering process that introduces two effective features such as Time Interval and ID Repetition Count. The experimental results show that MetaCAN consistently outperforms existing IDS solutions targetted the same datasets, making it a promising solution for securing the CAN bus in real-world vehicular environments.