Cyber-physical security: AI methods for malware/cyber-attacks detection on embedded/IoT applications

As the world becomes increasingly reliant on technology, the field of cybersecurity has gained paramount importance. The surge in the use of interconnected systems, particularly in the realm of autonomous vehicles, has escalated the risk of cyberattacks. Consequently, cyber-physical security has emerged as a critical area of research to address these concerns. The objective of this research was to delve into the field of cyber-physical security, focusing on the development of AI-based methods to detect cyberattacks on autonomous vehicles. Machine learning, a subset of artificial intelligence, has been extensively employed in cybersecurity to develop automated methods for detecting cyberattacks. Deep learning, in particular, has shown promising results in detecting anomalies and identifying cyberattacks. However, the complexity of these systems and the dynamic nature of the data generated by them pose significant challenges in implementing effective machine learning-based solutions. The research work presented in this thesis focused on the use of machine learning for cyber-attack detection in autonomous vehicles. A Simulink model was utilized to generate data and apply machine learning algorithms to detect cyberattacks. A Multi-layer Perceptron (MLP) model was selected as the final model, and the question of determining the number of layers and neurons in each layer was addressed using Neural Architectural Search (NAS). The final pipeline was written as clean code, and TensorFlow Lite was used to decrease the model size while maintaining accuracy. Through extensive experimentation involving 125 different model configurations, we found that 98 models achieved a Mean Absolute Error (MAE) of less than 3. Given the scale of the target variable, which ranges from 0 to 210, this level of error represents highly accurate predictions, with errors constituting only 1.42% of the target variable's range. The results of this research demonstrate that machine learning algorithms can be effectively used to detect cyberattacks in autonomous vehicles, providing a strong foundation for further research in this field. In conclusion, this research offers a comprehensive study of the field of cyber-physical security and the application of AI-based methods to detect cyberattacks in autonomous vehicles. The results underscore the potential of machine learning algorithms for detecting cyberattacks in autonomous vehicles and lay the groundwork for future research in this area.