AI for Cybersecurity : A Study on Machine Learning and DoS Attacks AI Robustness and Bypassing Detection Methods

Abstract: Cybercrime has increased for several years; both in volume andsophistication. When the capabilities of threat actors increase, techniques andtactics within cybersecurity also need to evolve. AI and machine learninghave potential to prevent and mitigate attacks. This report explores thepossible usage of machine learning for detection of DoS attacks, and furtherinvestigates the potential consequences of adversarial machine Learning. Weuse decision tree model that we train on publicly available DoS attack data.Then we use five computers to perform DoS attacks against a web server andcreate a machine learning model that attempts to detect the attacks based onthe attack's characteristics. In addition, we analyse the consequences ofadversarial machine learning with data poisoning. Our results show thepotential of using machine learning to detect DoS attacks and the dangers ofpoisoning attacks in this context.