ORADS: One Class Rule-Based Anomaly Detection System in Autonomous Vehicles
Anjanee Kumar, Tanmoy Kanti Das · IEEE Sensors Journal · 2025
Emergence of connected autonomous vehicles (CAVs) and driver assistant systems (ADS) is revolutionizing the transport sector, and people are expecting safe & convenient driving experience in future. Integration of information and communication technologies coupled with sensors and controllers enabled the growth of CAVs. Besides all obvious advantages, certain security vulnerabilities exist in the widely deployed communication standard known as the Controller Area Network (CAN), which is used for the transmission of information across different Electronic Control Units (ECUs) in CAVs. Several attacks have been demonstrated using those vulnerabilities, which can jeopardize the safe operation of CAVs. Though several anomaly detectors have been proposed recently, very few can detect such attacks in real-time without the use of additional computing resources. Here, we have proposed a design for a simple rule induction (SRI) framework to construct a real-time one-class anomaly detection system (ADS). The designed one-class rule-based ADS (i.e., ORADS in short) is capable of detecting unknown and adversarial attacks. The proposed approach has been verified using the CarChallenge 2020 dataset. It has been observed that ORADS has achieved F1 score of 0.92 for the test data of the final round when the classifier is trained with training samples given in the preliminary round of the challenge, and the said performance is significantly better than the stateof- the-art methods proposed elsewhere. Moreover, the ORADS requires 66.8 μs using a resource-constrained device like Raspberry Pi 3. Hence, ORADS can be implemented on consumer-grade conventional ECUs without any additional computing resources.