Federated learning-based misbehavior classification system for VANET intrusion detection

Federated learning-based misbehavior classification system for VANET intrusion detection

연구 분야: Infrastructure

학회: Journal of Intelligent Information Systems

In the domain of vehicular security, detecting misbehavior is crucial, particularly when faced with attacks that traditional cryptographic methods may overlook. Machine Learning (ML) and Deep Learning (DL) techniques offer a promising solution for identifying these sophisticated patterns of misbehavior. However, existing methods often rely on centralized systems, raising privacy concerns and exacerbating latency issues in the dynamic and scalable nature of vehicular environments. We develop a collaborative learning system that uses Federated Learning (FL) to identify misbehavior in vehicular networks in order to address these issues. Our approach aims to alleviate privacy concerns and mitigate latency problems by decentralizing the learning process across vehicles. We utilize the VeReMi extension dataset and conduct thorough evaluations across various client counts, employing different aggregation strategies such as FedAvg, FedProx, and FedYogi. Our experimental results highlight the accuracy of FL-based approach in detecting misbehavior within vehicular networks. These findings underscore the potential of FL to improve security while preserving privacy and meeting the stringent demands of real-time operations and scalability.