Optimal attention deep learning based in-vehicle intrusion detection and classification model on CAN messages

Area of Science:

• Cybersecurity
• Automotive Engineering
• Artificial Intelligence

Background:

• Controller Area Network (CAN) bus is crucial for automotive electronic control units (ECUs) but vulnerable to attacks.
• Intrusion detection systems (IDS) are vital for securing high-tech vehicles, with deep learning (DL) and machine learning (ML) showing promise.
• DL models require substantial data, posing challenges for CAN-based IDS.

Purpose of the Study:

• To present an Optimal Attention Deep Learning based In-vehicle Intrusion Detection and Classification (OADL-IVIDC) model.
• To enhance the security of CAN messages within vehicles.
• To address the data requirements of DL for in-vehicle IDS.

Main Methods:

• Data preprocessing to format input data.
• Utilizing an attention-based augmented long short-term memory (A-LSTM) model for in-vehicle IDS.
• Optimizing hyperparameters with the root mean square propagation (RMSProp) algorithm.

Main Results:

• The OADL-IVIDC model demonstrated superior performance compared to other techniques.
• The model effectively detects and classifies intrusions in CAN messages.
• Experimental validation was performed using a standardized car hacking dataset.

Conclusions:

• The proposed OADL-IVIDC model offers an effective solution for securing CAN bus communications in vehicles.
• The attention-based LSTM approach combined with RMSProp optimization enhances intrusion detection capabilities.
• This research contributes to advancing the safety and security of connected automobiles.