A Trust-based Client Selection Framework for Federated Learning in the Internet of Vehicles

Federated learning trains models on distributed data while preserving privacy, but its open and heterogeneous nature makes it vulnerable to malicious client updates that can compromise model integrity. Selecting trustworthy clients is crucial for secure and efficient operation, especially in dynamic environments like VANETs where vehicle mobility, communication reliability, and trustworthiness vary. We propose a trust-based client selection framework incorporating contextual information, reputation, and resource availability to mitigate the risks of faulty updates. Evaluated on the MNIST dataset, our approach demonstrates improved convergence, enhanced resilience against malicious clients, and superior performance compared to traditional random selection methods.