Misbehavior detection in intelligent transportation systems based on federated

Misbehavior detection represents a key security approach in vehicular scenarios to identify attacks that cannot be detected by traditional cryptographic mechanisms. In this context, the application of Machine Learning (ML) techniques has been widely considered to identify increasingly sophisticated misbehavior attacks. However, most of the proposed approaches are based on centralized settings, which could pose privacy issues, as well as an increased latency leading to severe consequences in the vehicular environment where real-time and scalability requirements are challenging. To address this issue, we propose a collaborative learning approach based on Federated Learning (FL) for vehicles' misbehavior detection. We use the reference misbehavior dataset VeReMi, which is re -balanced by applying the SMOTETomek technique. We carry out a thorough evaluation considering different balancing settings and number of nodes. The evaluation results overcome recent state-of-the-art approaches, with an overall accuracy of 93% using an optimized multilayer perceptron (MLP) for multiclass classification.