A secure and privacy preserved infrastructure for VANETs based on federated learning with local differential privacy

Advancements in Vehicular ad-hoc Network (VANET) technology have led to a growing network of interconnected devices, including edge devices, resulting in substantial data generation. The data generated by vehicles is subsequently shared with other devices, such as Roadside Units (RSUs). However, ensuring secure data sharing poses a significant challenge due to the potential risk of data breaches. Recently, Federated Learning (FL) has garnered substantial attention in the research community, enabling data owners to collaboratively learn a shared prediction model while retaining all their training data privately. However, traditional FL-based approaches are susceptible to inference and gradient leakage attacks. This paper presents a framework for private data sharing in VANETs using FL with local differential privacy. In the first layer, vehicles apply local differential privacy techniques to their data before sharing it with the RSU. The second layer is responsible for training model parameters at the RSU and updating the trained weights with the training server. To assess our system's performance, we evaluate it based on accuracy and simulation time for both local and global parameter sharing. Additionally, we measure each client's performance by calculating accuracy measures during each iteration. The experimental results demonstrate that our framework not only ensures security against inference and gradient leakage attacks but also exhibits superior efficiency compared to its counterparts.