Intrusion detection using metaheuristic optimization within IoT/IIoT systems and software of autonomous vehicles

Area of Science:

• Cybersecurity
• Machine Learning
• Automotive Engineering

Background:

• The increasing integration of Internet of Things (IoT) systems in vehicles introduces significant cybersecurity risks.
• Disruptions in automotive Controller Area Network (CAN) systems can lead to severe malfunctions, injuries, and fatalities.
• Ensuring vehicle safety and security is paramount, especially with the advancement towards autonomous driving.

Purpose of the Study:

• To investigate the efficacy of machine learning classifiers for detecting cyber intrusions in automotive CAN systems.
• To address the challenge of varying manufacturer standards by employing data-driven techniques.
• To improve the overall cybersecurity posture of connected vehicles.

Main Methods:

• Implementation and evaluation of Extreme Gradient Boost and K-Nearest Neighbor machine learning classifiers.
• Development of a modified metaheuristic optimizer for efficient parameter selection of the classifiers.
• Testing the proposed approach on a publicly available automotive CAN dataset.

Main Results:

• The best-performing machine learning models demonstrated an accuracy exceeding 89% in detecting cyber assaults.
• Statistical analysis confirmed the robustness of the optimizer's outcomes.
• Explainable AI techniques were used to identify key features influencing model performance.

Conclusions:

• Machine learning, particularly with optimized parameters, offers a viable solution for detecting cyber threats in CAN systems.
• The proposed data-driven approach enhances the security of automotive networks against evolving cyber threats.
• This research contributes to improving vehicle safety and reliability in the era of autonomous driving.