A Novel Machine Learning-Based Trust Management Against Multiple Misbehaviors for Connected and Automated Vehicles

Connected and Automated Vehicles (CAVs) are exposed to various threats in the dynamic, open and multi-domain network. Applying machine learning-based trust management for CAVs becomes imperative to capture complex features and adapt to dynamic situations. To deal with multiple misbehaviors and real traffic scenarios challenges, a trust management method for CAVs is proposed with data fusion, trust factor computation and two-tier trust prediction. First of all, three types of features on CAVs are analyzed, including spatio-temporal logic features, behavioral features and traffic flow features. Then, data fusion methods that combine beacon messages with map and detector data are proposed to enhance trust-related data. A multi-dimensional trust factor computation approach is then introduced using statistical methods. Finally, per-minute and multi-minute machine learning-based trust prediction methods are performed at the node level using the computed trust labels from each beacon. The results showed the effectiveness and real-time capability of the data fusion process, as well as the completeness of the trust factor computation. The trust prediction results showed both high performance at the per-minute level with models like XGBoost and further improved performance at the multi-minute level with deep learning models.