Detecting Cyber Attacks In-Vehicle Diagnostics Using an Intelligent Multistage Framework.

Area of Science:

• Cybersecurity
• Automotive Engineering
• Machine Learning

Background:

• Modern vehicles are increasingly vulnerable to cyber threats due to advanced technology.
• Existing Intrusion Detection Systems (IDS) face challenges with high false acceptance rates and network infrastructure modifications.
• Electronic Control Unit (ECU) limitations and placement concerns impact IDS effectiveness.

Purpose of the Study:

• To propose a novel multistage framework for detecting abnormalities in vehicle diagnostic data.
• To enhance vehicle security without significant changes to existing network infrastructure.
• To achieve high detection accuracy and a low false acceptance rate for vehicle intrusion detection.

Main Methods:

• Developed a multistage framework utilizing diagnostic specifications.
• Employed a stacking ensemble of various machine learning models for anomaly detection.
• Validated the framework against KIA SOUL and Seat Leon 2018 datasets.
• Evaluated performance against unseen point and period anomaly attacks.

Main Results:

• Achieved high accuracy: 99.21% for Seat Leon 2018 and 99.22% for KIA SOUL.
• Demonstrated a low false acceptance rate: 0.003% for Seat Leon 2018 and 0.005% for KIA SOUL.
• Reported a high detection rate (DR): 99.63% for Seat Leon 2018 and 98.59% for KIA SOUL.

Conclusions:

• The proposed multistage IDS framework is superior and robust in detecting vehicle network anomalies.
• The system effectively identifies various anomaly types with excellent performance metrics.
• This approach offers a promising solution for enhancing automotive cybersecurity.