Securing the CAN bus using deep learning for intrusion detection in vehicles.

Area of Science:

• Cybersecurity
• Automotive Engineering
• Machine Learning

Background:

• The Controller Area Network (CAN) bus is critical for vehicle communication but lacks security, leaving Intelligent Transportation Systems (ITS) vulnerable to cyberattacks.
• Attacks like Denial of Service (DoS), fuzzing, impersonation, and spoofing pose significant risks to vehicle safety and data integrity.

Purpose of the Study:

• To evaluate the efficacy of deep learning methods for detecting intrusions within the CAN bus network.
• To assess the performance of various Recurrent Neural Network (RNN) architectures in identifying automotive cyber threats.

Main Methods:

• Utilized three datasets: Car Hacking, Survival Analysis, and OTIDS for training and testing deep learning models.
• Explored Recurrent Neural Network (RNN) variants, including Long Short-Term Memory (LSTM), Gated Recurrent Unit (GRU), and VGG-16, to analyze temporal and spatial features of CAN messages.
• Implemented Bi-LSTM for enhanced sequence analysis by processing data in both forward and backward directions.

Main Results:

• LSTM achieved 99.89% accuracy in binary classification tasks for intrusion detection.
• VGG-16 demonstrated 100% accuracy in multiclass classification scenarios.
• The models effectively identified anomalies and cyber threats within the CAN bus data.

Conclusions:

• Deep learning techniques, particularly LSTM and VGG-16, show significant potential for bolstering the security of Intelligent Transportation Systems (ITS).
• These methods offer a robust solution for detecting and mitigating cyberattacks on CAN bus networks, improving overall vehicle resilience.