Digital Twin-Enabled Federated Misbehavior Detection Model for Internet of Vehicles Networks

The concept of federated learning enhances the security and privacy of Internet of Vehicles (IoV) networks while accelerating the development of more intelligent, secure, and efficient vehicular applications. Federated learning-based approaches allow the training of machine learning models directly on the IoVs without transferring the sensitive information to a Mobile Edge Computing (MEC) server. One persisting drawback in these approaches is that malicious IoVs may deliberately engage in malicious behaviors to manipulate the edge model training and reduce the trustworthiness of the system. To overcome this challenge, we propose a digital twin-enabled federated misbehavior detection model to collaboratively detect malicious behaviors in IoV networks without sharing sensitive data. Combining digital twin with IoV networks, we analyze real-world vehicle behavior and investigate various attack scenarios. We develop a federated misbehavior detection algorithm, which continuously learns and improves system performance by identifying malicious activities of potential attacks. This algorithm allows each IoV to process data locally and contribute to an edge model for a more robust and secure misbehavior detection model. Our simulation results show that the proposed federated learning model with the XGBoost algorithm detects attacks with a maximum accuracy of 99%, while also ensuring the trustworthiness of the data.