Machine learning-based identification of cybersecurity threats affecting autonomous vehicle systems

Abstract

With the advancement of humanity, transportation and trade activities have increased, leading to the development process of basic land vehicles as more than physical power became necessary. Hand tools were developed with the invention of the wheel, followed by animal-powered vehicles, and then steam engine technology. After the advancement of electromechanical technologies, today's modern vehicles have been developed. Those who used these vehicles thought of transferring control from the human to autonomous driving systems to solve their safety and comfort problems. Today, instead of fully autonomous systems targeted for the future, autonomous driving support systems have been developed. Although these systems aim to increase the safety and comfort of passengers, they can become an easy target for malicious people due to network technologies and remote connection features. The most effective method of protection from these attackers is to conduct vulnerability analysis against newly emerging threats for the systems we use and to rectify identified vulnerabilities. In this research paper, the weaknesses of wireless communication towards remote connection usage of the mini electric autonomous vehicle were investigated, which we developed and produced its mechanics, electronics, and software. In this context, a test environment was created, and the problems and threats in autonomous driving technology were revealed through attacks (Deauth Attack, DoS, DDoS and MitM) made on the test environment. Following the exposed vulnerabilities, studies were conducted for the detection of these attacks using Artificial Intelligence. In the study, different algorithms were used to detect the attacks, and random forest algorithm successfully detected 96.1% of attacks. The main contribution to the field of cybersecurity in autonomous vehicles by providing effective solutions for threat identification and defense.