In-Vehicle Networks Security Using Transfer Learning Approach Against AI-Generated Cyberattacks

Autonomous vehicles (AVs) are revolutionizing transportation by offering unparalleled efficiency and reshaping the way people travel. However, this technology also raises concerns due to its susceptibility to unwanted control by unau-thorized users, which can result in significant damage. The importance of in-vehicle networks has gained attention from researchers who are working to protect Controller Area Networks (CAN), an integral part of AVs. Despite the presence of various techniques proposed by researchers to protect these CANs, there are still many issues, such as addressing new, unseen attacks and protecting against AI-based attacks. To address these challenges, this study proposes an approach that can effectively handle diverse attacks, including AI-based attacks, using a machine learning approach. We utilize a dataset from our previous study, which consists of hybrid traffic comprising original attack traffic and AI-based attack traffic generated by CTGAN. The proposed approach incorporates the concept of transfer learning, where we leverage a trained CNN to extract valuable feature patterns. These CNN-based features are then used to train machine learning algorithms. Our results demonstrate that Random Forest models achieved significant success, with an accuracy score of 0.92 and very low computational training and testing times of 78.5501 and 2.2262, respectively.