Investigating Vulnerabilities and Potential Security Threats In Current Synthetic Data Generation And Usage In The Automotive Domain

Abstract

In the rapidly evolving automotive industry, ensuring robust cybersecurity measures is crucial for the safe and reliable operation of modern vehicles. This thesis investigates the generation of synthetic Controller Area Network (CAN) data using Generative Adversarial Networks (GANs) to enhance the security and privacy of automotive systems. The study focuses on two models: Simple GAN and WGAN-GP, evaluating their effectiveness in generating synthetic CAN data that closely mimics real CAN data. The research begins with an in-depth analysis of the CAN protocol and its vulnerabilities, particularly to Denial of Service (DoS) attacks. By utilizing the CAN Dataset for Intrusion Detection (OTIDS), the study generates synthetic CAN data under normal and DoS attack scenarios. The generated data is evaluated using various metrics, including mean and variance comparison, Dynamic Time Warping (DTW), Kolmogorov-Smirnov (KS) test, and classification accuracy. Visual inspections and RandomForest classification analysis are also conducted to assess the indistinguishability and realism of the synthetic data. The findings indicate that both Simple GAN and WGAN-GP models effectively replicate the statistical properties and temporal dynamics of real CAN data, with WGAN-GP showing slightly better performance in handling DoS attack data. The synthetic data's utility for cybersecurity applications, such as intrusion detection systems (IDS), is demonstrated, highlighting its potential for rigorous security testing while ensuring data privacy. The thesis concludes with a discussion of future research directions, including enhanced GAN architectures, extended evaluation metrics, real-world implementation, and cross-domain applications. The integration of GANs with Large Language Models (LLMs) for automated requirement analysis, effort estimation, and continuous learning is also explored. These advancements are aimed at addressing evolving challenges and opportunities in automotive cybersecurity, ensuring a balance between innovation and security