Intrusion Detection in Controller Area Network Bus Using Deep Neural Network

Today's smart cars have hundreds of embedded systems or Electronic Control Units (ECUs) to provide luxury and comfort to customers. These features are all connected with the Controller Area Network (CAN) bus. CAN is responsible for the communication between these ECUs. Wireless communication among CAN’s ECUs also makes it remarkably vulnerable. Once the CAN system becomes compromised, the hacker can send commands and control ECUs. This unguarded system is dangerous and can result in fatal damage to those in the vehicle and the environment around it. We design an Intrusion Detection System (IDS) based on Deep Neural Network to counter three critical attack categories, namely Denial of Service attack, Fuzzy attack, and Impersonation attack. The IDS learns to classify incoming packets from the CAN bus by processing the CAN Identifier and relevant data information. The performance of the IDS is calibrated by two datasets that include an available online dataset and synthesized intrusion messages for real time datasets collected by us. Evaluation results show that our IDS effectively detects all three attack types for both types of datasets and verifies that it outperforms all existing similar approaches in terms of accuracy, scalability, and low false alarm rates