Machine Learning-Based Intrusion Detection System for Controller Area Networks

The automotive industry continues to innovate at an exponential rate to provide a safer and more efficient experience for consumers. Autonomous vehicles and Vehicle-to-Everything technologies are at the forefront of defining the future of transportation. Enabling vehicles to connect to various services has exposed critical in-vehicle networks such as the Controller Area Network (CAN) to potential exploitation by adversaries. In its standard form, the CAN bus suffers from multiple vulnerabilities such as limited bandwidth and lack of authentication. Attacks can be initiated through physical and wireless mediums, exploiting diagnostic interfaces, Bluetooth and infotainment systems to compromise the confidentiality, integrity and availability of data communication within vehicles. In this paper, a holistic, comprehensive, Machine Learning-Based intrusion detection system for the CAN bus is proposed to secure the critical in-vehicle network. The proposed system is modular, scalable and can be adapted to the ever-changing threat landscape of cyber vehicle attacks. On an unseen testing dataset, our system achieved 100% accuracy in protecting against denial of service and multiple impersonation injection attacks, as well as 95.67% accuracy of fuzzy injection attacks.