Performance Analysis of Various Machine Learning Algorithms in the Classification of Intrusion Traffic in VANET Communication

Abstract

Modern vehicles are more than just machines. In order to implement various functionalities and carry out actions, it utilises numerous electronic control units connected via  intra vehicle networks.As a promising technology for VANET communication systems, Intelligent Transportation Systems (ITS) has the potential to widen security loopholes by connecting all of the vehicle networks together. To meet VANET's various security and privacy requirements, there are many challenges to overcome.The Controller Area Network (CAN) standard is used to transmit data reliably and efficiently in a realtime vehicle network. An attacker can easily cause system failures by injecting any message  into the CAN because the protocol sends messages in broadcast mode. ECU security and the CAN bus security are the most important aspects of in-vehicle security, so an intrusion detection system (IDS) that uses machine learning methods to detect malicious cyber-attacks is developed to make sure this doesn't happen in the future. To detect Dos and Fuzzy attacks and classify intrusion traffic, our system employs four supervised algorithms: K- Nearest Neighbor, Support Vector Machines, Nave Bayes, and Decision Tree.There are four performance metrics used: Accuracy, F-score, Precision, and Recall. For the classification of legitimate and attacked traffic in a VANET environment, our proposed system suggests the best algorithm based on the performance metrics listed above. In our experiments, we compare the performance of various machine learning algorithms using different ratios of training to testing data. According to our findings, data splitting is critical when it comes to evaluating the performance of machine learning algorithms.