A SCALABLE INTRUSION DETECTION ARCHITECTURE FOR AUTONOMOUS VEHICLE NETWORKS USING ENSEMBLE LEARNING

The rapid adoption of AVs has increased the need for robust cybersecurity mechanisms to safeguard in-vehicle communication networks, such as the CAN bus. Intrusion detection systems (IDS) have emerged as a critical line of defense, yet their practical deployment in AV environments faces challenges of scalability, latency, and accuracy. This study proposes a scalable intrusion detection architecture that integrates EL methods with a layered framework designed for real-time vehicular cybersecurity. The system incorporates preprocessing, feature selection, and ensemble classification to detect denial-of-service, spoofing, fuzzy, and gear spoofing attacks alongside normal traffic using a large-scale dataset of more than 16 million CAN traffic records. Experimental results show that Random Forest and XGBoost deliver superior performance, achieving detection rates above 99.9% in binary classification and over 98.6% in multi-class classification, while maintaining reduced training times when feature selection is applied. By leveraging efficient model integration and streamlined data preprocessing, the proposed architecture emphasizes scalability, making it suitable for large-scale vehicular deployments. This work contributes both algorithmic and system-level insights by demonstrating that EL models can be effectively embedded in a scalable IDS framework. The findings provide practical guidance for designing robust, real-time IDS solutions for AV networks.