Lightweight Detection of Abnormal Behavior Based on Semantic Features of In-Vehicle Networks

Security has been an increasing problem for intelligent connected vehicles (ICVs). Abnormal behavior of in-vehicle networks is the key for threat identification and risk prevention of ICVs. Most of the abnormal behaviors are statistical features reflecting attack models. These features are simple and easy to extract and use for lightweight detection of abnormal data. In the detection process, most attacks are detected with as lightweight overhead as possible to detect abnormal behavior for in-vehicle networks. Aiming at the abnormal behavior features of ICVs for different attacks, this paper proposes an abnormal behavior detection algorithm of in-vehicle networks based on semi-supervised learning. We label the abnormal data and test the unlabeled data to verify the detection performance of the proposed method, and also present the possibility of detection for unknown attacks.