Misbehavior Detection Based on Support Vector Machine and Dempster-Shafer Theory of Evidence in VANETs

Vehicular ad hoc networks (VANETs) support safety and comfortable driving through frequent information exchange among intelligent vehicles. As an open access environment, VANETs are vulnerable to security threats, such as electronic attack and privacy disclosure. In this paper, we propose a misbehavior detection mechanism based on a support vector machine (SVM) and Dempster-Shafer theory (DST) of evidence to resist false message attack and message suppression attack. The proposed mechanism includes data trust model and vehicle trust model. The data trust model uses an SVM-based classifier to detect false messages based on message content and vehicle attributes. The vehicle trust model consists of a local vehicle trust module and a trust authority (TA) vehicle trust module. The local vehicle trust module uses another SVM-based classifier to evaluate whether the vehicle is credible based on the behavior of the vehicle in terms of message propagation. Then, the TA vehicle trust module uses DST to aggregate multiple trust assessment reports about the same vehicle and derives a comprehensive trust value. Simulation results show that Gaussian kernel best fits our models compared with other functions. In addition, the true positive rate of our data trust model is higher than the model based on back propagation neural network. Moreover, our two models are more robust than basic majority voting, weighted voting, and Bayesian inference in terms of true positive rate under various scenarios.