Protecting Automotive Controller Area Network: A Review on Intrusion Detection Methods Using Machine Learning Algorithms

To manage the future requirements for comfortable, safe, and low-carbon driving, the in-vehicle network is undergoing rapid developments. Despite the huge changes in its architecture, we claim that Controller Area Network (CAN), which is developed for more than 35 years, would continue to play a critical role in ensuring the safety of vehicles. However, its intrinsic vulnerability to cyber-attack becomes one of the biggest challenges since vehicles are no longer isolated. The intrusion detection approach draws much attention due to its simplicity and efficiency in protecting in-vehicle CAN bus. In this chapter, we first provide an introduction about how the in-vehicle network evolves. The critical role of CAN for current and future in-vehicle network is emphasized. Next, we describe intrusion detection approaches that exploit machine learning algorithms in detail. The introduction is taken from four aspects according to the domain knowledge of CAN used for intrusion detection methods, which are semantics-based methods, literal-based methods, timing-based methods, and signal characteristics-based methods respectively.