Securing CAN Bus in Connected and Autonomous Vehicles Using Supervised Machine Learning Approaches

Abstract

Connected and Autonomous Vehicles (CAVs) are becoming a promising solution in Intelligent Transportation Systems (ITS). Despite these advancements, vehicles still use a Controller Area Network (CAN) bus system to communicate among the different electronic control units (ECUs). While very efficient for message transmission, the CAN bus has significant security holes as the network is unsegmented, unencrypted, and lacking authentication. For these reasons, it is necessary to implement another security barrier for vehicles to continue utilizing CAN. Due to a large amount of data and different attack patterns, intrusion detection for CAN bus attacks is still challenging. This paper proposes the use of time intervals as a new feature for intrusion detection on the CAN bus. A comparison of supervised machine learning algorithms, including Random Forest (RF), K-Nearest Neighbor (KNN), and XG boost, is conducted with different sets of features. Experimental results show that the RF classifier, with an accuracy of 97%, outperforms both KNN and XG boost for detecting attack messages in the CAN intrusion dataset. In addition, the results indicate that the inclusion of time intervals in the feature set improved the accuracy of RF by 3%.