Trust-based secure federated learning framework to mitigate internal attacks for intelligent vehicular networks

Vehicular Ad-hoc Networks (VANETs) are increasingly vulnerable to internal threats, such as Sybil attacks and insider malicious behaviors, which exploit the network from within. Existing security measures often fail to detect these sophisticated threats, leaving the network exposed to significant risks. This paper presents a novel trust-based secure Federated Learning (FL) framework specifically designed to address these internal threats. The proposed framework introduces several technical innovations such as a real-time behavioral consistency monitoring system that detects anomalies in node behavior over time, a trust score decay mechanism to prevent long-term exploitation by malicious nodes and a decentralized trust validation process that ensures robustness against Sybil attacks. By incorporating these trust evaluations, the federated learning model selectively aggregates learning updates, prioritizing those deemed reliable, thereby enhancing the overall robustness of the detection model. Extensive experimental simulations were conducted on various datasets, and four types of attacks were analyzed. The results demonstrate significant improvements in detection rates of internal attacks with minimal impact on network performance. The precision and accuracy of the proposed model are improved by 98.5% and 98.8%, respectively. The network throughput and attack detection rates improved significantly for internal threats over the existing models.