Machine Learning Based Intrusion Detection System for Internet of Vehicles

Vehicular networks, especially the Controller Area Network (CAN) bus, are becoming more susceptible to hacks that could imperil safety-critical functions due to the growing connected automobiles. In order to mitigate such threats, Intrusion Detection Systems (IDSs) are essential. Nevertheless, it is still difficult to develop an IDS that is both extremely accurate and resilient. Our work suggests a thorough intrusion detection methodology for CAN bus security that is based on machine learning. A non-linear Decision Tree and baseline classifiers like Naïve Bayes, Linear Discriminant Analysis, Logistic Regression, and K-Nearest Neighbors are assessed alongside ensemble-based models like Random Forest, Extra Trees, and Extreme Gradient Boosting (XGBoost). Stratified data splitting and suitable feature preprocessing are used in extensive tests on a large-scale CAN intrusion dataset “Car Hacking” with millions of samples. For performance evaluation Precision, Accuracy, F1-score, Recall, confusion matrix analysis, and Receiver Operating Characteristic - Area Under the Curve are used. Experimental results show that ensemble-based tree models provide near-perfect detection accuracy and excellent class separability, greatly outperforming baseline techniques. For safety-critical car intrusion detection, Random Forest and Extra Trees in particular produce minimal false negatives and few false positives, which makes them ideal. The results showed how well ML models works for IOV security and offer justification for using Random Forest as a trustworthy backbone IDS model.