COMPARATIVE STUDY OF ENSEMBLE LEARNING ALGORITHMS FOR INTRUSION DETECTION IN AUTONOMOUS VEHICLE NETWORKS

The rapid advancement of AVs has heightened the need for robust cybersecurity measures to protect in-vehicle communication networks. Intrusion detection systems (IDS) play a critical role in identifying and mitigating malicious activities. This study presents a comparative evaluation of six widely used ensemble learning algorithms—Random Forest, XGBoost, LightGBM, CatBoost, Extra Trees, and AdaBoost applied to a large-scale autonomous vehicle dataset containing over 16 million records. Each model was trained to detect four major attack types, i.e., denial-of-service, spoofing, fuzzy, and gear spoofing, alongside normal traffic. Performance was assessed under both binary (attack vs. normal) and multi-class (specific attack categories) classification settings using accuracy, precision, recall, and F1-score. Results show that Random Forest and XGBoost consistently outperformed the other classifiers, achieving near-perfect binary detection rates of 99.99% and multi-class accuracy above 98.6%. LightGBM and Extra Trees also performed well, whereas CatBoost and AdaBoost lagged in terms of precision for certain attack categories. This study provides practical insights into the comparative strengths of ensemble methods for vehicular IDS, highlighting random forest and XGBoost as the most reliable models for large-scale autonomous vehicle cybersecurity applications