Enhancing Intrusion Detection in The Internet of Vehicles: An Ensemble and Optimized Machine Learning Approach

In contemporary intelligent transportation systems, particularly within autonomous vehicles and connected cars, a growing array of functionalities has been integrated through interactions with other vehicles, smart devices, and infrastructure. However, the increasing connectivity in the Internet of Vehicles (IoV) has concurrently heightened the risk of network attacks. To fortify IoV systems against these threats, intrusion detection systems (IDSs) utilizing machine learning techniques prove effective in identifying malicious network attacks. This study proposes a novel optimized ensemble IDS employing soft voting for the detection of cyber threats in both in-vehicle networks and external vehicular networks within modern intelligent transportation systems. The system framework is established on three machine learning algorithms (logistic regression, random forest, and decision tree) optimized by Bayesian optimization, creating an integrated structure capable of efficiently detecting intrusions from both in-vehicle and external networks. The proposed framework undergoes evaluation using two benchmark datasets: the CICIDS2017 dataset for external network communication and the Car-Hacking dataset for in-vehicle communication. The findings indicate that the proposed framework attains an accuracy of 99.997% on the Car-Hacking dataset and 99.515% on the CICIDS2017 dataset, surpassing the performance of existing intrusion detection systems.