A Machine Learning Approach for Detecting Internal Position Falsification Attacks in VANET Environments

Abstract

As decentralized and self-organizing wireless networks, Vehicular Ad-hoc Networks (VANETs) enable direct communication among vehicles and infrastructure, forming a key component of Cooperative Intelligent Transportation Systems (C-ITS). However, their open and dynamic nature makes them vulnerable to severe security threats, particularly internal attacks carried out by legitimate but malicious nodes. A notable internal threat is position falsification, where a malicious vehicle transmits false location data in Basic Safety Messages (BSMs), misleading nearby vehicles and potentially causing traffic disruptions and accidents. To combat these threats, Misbehavior Detection Systems (MDSs) have been developed to identify abnormal vehicle behavior through contextual and data-driven analysis. This study evaluates several machine learning algorithms on the VeReMi dataset, addressing both binary and multi-class classification attacks. The experiments achieved high performance metrics, including F1 scores of up to 99% in multi-class scenarios through effective code optimization and hyperparameter tuning. Methods like Bagging with RF, and Gradient Boosting showed excellent accuracy, while K-Nearest Neighbors (KNN) excelled with low execution times. Overall, RF proved to be one of the most effective classifiers, offering a scalable approach to detect position falsification and VANET attacks.