A Robust Misbehavior Detection System for Cooperative Driving Network

Cooperative Intelligent Transportation Systems (c-ITS) stand as visionary pillars in shaping the future landscape of vehicular networks. These systems hinge on the seamless orchestration of Vehicle-to-everything (V2X) communication, fostering the seamless exchange of vital information between vehicles and the underlying infrastructure. Within this intricate ecosystem, vehicles collaborate harmoniously, disseminating V2X messages throughout the network. However, the transmission of deceptive or inaccurate information by rogue vehicles has the potential to endanger road safety. Misbehavior Detection systems (MBS) perform plausibility and consistency checks on the received V2X information or consider machine learning models training while ignoring the variability of the traffic over time. In this paper, we introduce a stacked meta-learning model for misbehavior detection, leveraging time-series V2X data to distinguish between genuine and aberrant exchanged information. Our proposed approach integrates predictions from various probabilistic learning models, incorporating them into a CNN-LSTM meta-learner to weigh the outputs of the probabilistic classifiers for accurate misbehavior detection. Our method demonstrates its ability to identify a wide range of faults and attacks, including those associated with rare types of attack, by learning the inherent data distribution and essential characteristics of network traffic. Simulation results reveal that the proposed model not only outperforms prominent MBS in the literature by accurately identifying various forms of misbehavior within cooperative driving systems but also reduces the overall execution time of aberrant behavior detection.