Countering Attacks in IN-Vehicle Network: An Evaluation of Machine Learning Algorithms

초록

Cyber-security for IN-Vehicle network protocols is a high-priority research area in the automotive space. Electronic Control Units (ECUs) in IN-Vehicle networks are connected via the CAN protocol to facilitate reliable information dissemination. Nevertheless, devices connected to this protocol, and the protocol itself are susceptible to attack as it lacks data authentication mechanism for detecting legitimate and illegitimate traffic. A variety of Machine Learning (ML) approaches for mitigating this problem have emerged. The work presented an investigation into the classification power of various ML models using a public CAN bus dataset. In addition, a comparative analysis of the most used and standardized ML techniques. Based on the analysis of the results of the various investigations, the "Fine Tree", "Fine Gaussian SVM", and "Fine KNN" are the most suitable for resolving security issues in CAN Bus using the accuracy and minimum classification error of the various ML models as evaluation criteria.