Adaptive Intrusion Detection in CAN-Based Vehicular Networks Using Transfer Learning for Evolving Threats

This is the CAN-based system security, which is constantly under serious threats due to increasing vehicular network connectivity through sophisticated cyber-attacks. In this paper, we propose an online reconfigurable IDS which uses TL methods to adapt to new attack patterns with minimal retraining. That would allow refining the pre-trained models on the specialized car hacking dataset to detect most of the known and new attacks efficiently, ensuring high detection accuracy while keeping the computation cost low. Dynamic reconfigurability assures protection in an ever-evolving threat landscape. Extensive experiments on real-world CAN datasets validate the effectiveness of the proposed approach, achieving an overall detection rate of over 99% for different types of attack classes. The approach here is toward demonstrating the potential of TL for enhancing adaptability, efficiency, and accuracy in improving connected vehicle IDSs through a sound solution to secure automotive systems against the emergence of cyber threats.