An Effective Abnormal Behavior Detection Approach for In-Vehicle Networks Using Feature Selection and Classification Algorithm

Abnormal detection has become an essential method of security protection in automotive. In order to meet the compatibility between electronic control units (ECUs), the data transmitted in controller area network (CAN) bus need to obey different protocols and specific communication rules. Generally, these rules can be learnt through statistics and used in the abnormal detection of in-vehicle networks. However, satisfactory abnormal detection performance is hard to guarantee when the in-vehicle network communication rules is relatively simple. To improve the detection performance comprehensively, this paper chooses the classification algorithm to carry out anomaly detection. Considering the particularity of in-vehicle networks, this paper proposes a classification algorithm based on the feature vectors of CAN bus packets. Combining with the feature vectors, the convolutional neural network (CNN) algorithm is used to realize high detection performance for in-vehicle networks. Moreover, the real vehicle experiment has verified the satisfactory results for the proposed classification algorithm.