New Features for Position Falsification Detection in VANETs using Machine Learning
Seçil Ercan, Marwane Ayaida, Nadhir Messai
Abstract
Misbehavior detection in VANETs is critical to provide security and road safety since the communication between vehicles is directly affected by attackers’ messages. A new misbehavior detection approach is presented and compared with previous approaches. The estimated angle of arrival and the estimated distance using path loss model are proposed as new features to use for detection mechanism. When an attacker sends a message, it can produce false position values. Therefore, using a set of suitable features about the relation between sender and receiver will help to detect attackers. Machine learning techniques, Random Forest and k-Nearest Neighbor, are implemented to classify a vehicle as an attacker or not by considering the proposed features. The performance of the proposal is evaluated using a public dataset and easily compared to others. The results show that the proposed approach increases the performance of misbehavior detection in terms of classification evaluation metrics.