A lightweight intrusion detection approach for CAN bus using depthwise separable convolutional Kolmogorov Arnold...

Area of Science:

• Cybersecurity
• Artificial Intelligence
• Automotive Engineering

Background:

• Modern vehicles increasingly rely on connected and autonomous systems, heightening cybersecurity risks.
• Existing intrusion detection systems (IDS) struggle with imbalanced datasets and high computational costs, hindering practical automotive application.

Purpose of the Study:

• To develop a robust and computationally efficient IDS for automotive cybersecurity.
• To address data imbalance issues in detecting vehicle network attacks.
• To enhance the practical deployment of IDS in connected and autonomous vehicles.

Main Methods:

• Utilized spectral normalization Generative Adversarial Networks (GAN) to synthesize anomalous vehicular network data, balancing datasets across attack types and normal traffic.
• Developed a lightweight Depthwise Separable Convolutional Kolmogorov-Arnold network (DSC-KAN) incorporating the Kolmogorov-Arnold (K-A) theorem for efficient classification.
• Evaluated the proposed IDS against existing methods using key performance metrics.

Main Results:

• The proposed IDS achieved a balanced data distribution for four attack categories and normal traffic.
• The DSC-KAN model demonstrated superior accuracy and significantly improved computational efficiency compared to existing methods.
• Experimental results validated the effectiveness and practicality of the developed IDS solution.

Conclusions:

• The novel IDS approach effectively addresses limitations of current systems in automotive cybersecurity.
• The combination of GAN for data synthesis and DSC-KAN for classification offers a promising solution for real-world vehicle network security.
• This research contributes to safer and more reliable connected and autonomous driving technologies.