Misbehavior Detection in VANET's: Six Different Approaches for VeReMi Dataset Using Machine Learning Algorithms

In the ever-evolving landscape of autonomous vehicle technology, ensuring the safety and reliability of selfdriving cars has become paramount. This effort has resulted in the investigation of novel techniques to improving the safety and reliability of autonomous cars. This research looks towards detecting misbehavior in Vehicular ad-hoc network (VANET's) by proposing six different approaches to apply on the VeReMi data, and evaluate the performance of all nine state-of-the-art ML algorithms over binary classification on for five different attack types. The 10-fold cross-validated XGBoost (XGB) performed better over other 8 state-of-theart ML algorithms in addressing a wide range of attack scenarios despite class imbalance, with average F1-sores of: 79.234% for Approach-1, 80.16% for Approach-2, 78.232% for Approach-3, 79.262% for Apporach-4, 78.234% for Approach-5, and 78.288% for Approach-6 on all five attacks, leading to conclude that Approach-2 is more robust and secure VANET systems.