Multi-attack and multi-classification intrusion detection for vehicle-mounted networks based on mosaic-coded...

Area of Science:

• Cybersecurity
• Automotive Engineering
• Machine Learning

Background:

• The increasing connectivity of vehicles via the Internet of Vehicles (IoV) exposes them to significant external network attack risks.
• Controller Area Network (CAN) bus, a widely used in-vehicle network, faces complex and diverse attack vectors.
• Existing intrusion detection systems often struggle to detect multiple simultaneous attack types, necessitating advanced solutions.

Purpose of the Study:

• To develop a single, unified intrusion detection model capable of identifying multi-type cyberattacks on vehicle CAN buses.
• To enhance the detection accuracy and multi-classification ability for various combinations of CAN bus attacks.

Main Methods:

• Proposed a Mosaic-coded convolution neural network (CNN) for intrusion detection.
• Created a Mosaic-like 2D data grid from 1D CAN IDs to preserve temporal features for the CNN.
• Employed an autoencoder for data dimensionality reduction to optimize model complexity.

Main Results:

• The Mosaic-CNN model demonstrated high and stable multi-classification ability in detecting various attack combinations.
• Effectively identified all tested attack types and their combinations on the CAN bus.
• Dimensionality reduction using autoencoders contributed to a less complex model.

Conclusions:

• The proposed Mosaic-coded CNN offers an effective solution for detecting complex, multi-type cyberattacks on vehicle CAN buses.
• This approach significantly improves the capability of a single model to handle diverse and combined intrusion scenarios.
• The method provides a robust framework for enhancing the security of connected vehicles.