A Lightweight Intrusion Detection System with Dynamic Feature Fusion Federated Learning for Vehicular Network...

Area of Science:

• Cybersecurity
• Artificial Intelligence
• Automotive Engineering

Background:

• Autonomous vehicles face increasing cybersecurity risks due to complex sensors and ECUs.
• Controller Area Network (CAN) lacks inherent security, making it vulnerable to attacks.
• Traditional Intrusion Detection Systems (IDS) struggle with the dynamic nature of modern vehicular threats.

Purpose of the Study:

• To propose a lightweight vehicular network intrusion detection framework using Dynamic Feature Fusion Federated Learning (DFF-FL).
• To enhance the detection of complex cyber threats in autonomous vehicle networks.
• To ensure data privacy and suitability for edge device deployment.

Main Methods:

• A two-stream architecture combining a transformer-augmented autoencoder for feature extraction and a CNN-LSTM-Attention model for pattern preservation.
• Dynamic fusion of node feature representations via a transformer attention module for cross-node interaction in heterogeneous data.
• An adaptive weight adjustment mechanism based on node performance to improve global model robustness.

Main Results:

• Achieved over 99% F1 score on the CAN-Hacking dataset.
• Demonstrated a lightweight system with only 1.11 MB memory and 81,863 trainable parameters.
• Maintained low computational overhead while ensuring data privacy.

Conclusions:

• The DFF-FL framework offers a robust and efficient solution for vehicular network intrusion detection.
• The system's lightweight nature and high accuracy make it ideal for edge deployment in autonomous vehicles.
• This approach effectively addresses the limitations of traditional IDS in complex, dynamic vehicular environments.