Machine Learning Techniques for IoT Intrusions Detection in Aerospace Cyber-Physical Systems

Abstract

Aeronautical systems are no longer traditional masterpieces of autonomous mechanical engineering. Today, they are characterized by many intelligent technologies that include sensors, wireless standards and data analysis tools. Known as Aerospace Cyber-physical Systems (CPS), these CPSes are undergoing a massive transformation to increase the safety, efficiency and reliability of their operations. The physical system has created the Internet of Things IoT by integrating sensors, controllers and actuators. Nevertheless, the cyberspace of these aerospace CPSes offers many opportunities for malicious actors who threaten the security and privacy of vehicles/aircraft and their applications. Unprotected or poorly protected systems can easily be exploited for malicious purposes. Indeed, aerospace CPSes are always under threat from an increasing number of cyber-attacks through sensory or wireless channels, hardware, software or actuators. Recently, due to the significant advances and impressive results of machine learning techniques in the fields of image recognition, natural language processing and speech recognition for various long-standing artificial intelligence tasks, there has been a great interest in applying them to intrusion detection in the field of cybersecurity. In this chapter, we present different machine learning techniques for IoT intrusion detection in aerospace cyber-physical systems. The application of machine learning for cybersecurity in IoT requires the availability of substantial data on IoT attacks, but the lack of data on IoT attacks is a significant problem. In our study, the Cooja IoT simulator was used to generate high fidelity attack data in IoT 6LoWPAN networks. The efficient network architecture for all machine models is chosen based on comparing the performance of various network topologies and network scenarios. The experimental results show that Machine learning models for intrusion detection give better results by more than 99% in terms of accuracy, efficiency and detection rate. Also, it requires a low energy consumption overhead and memory, which proves that the proposed models can be used in constrained environments such as IoT sensors.