Machine Learning-Based Intrusion Detection for Denial-of-Service De-Authentication Attacks on Drones

As Unmanned Ariel Vehicles (UAV) or drones are being more widely adopted for various purposes in many sectors, their safety and security is becoming an increasing concern. Cyber-attacks on drones such as denial-of-service (DoS) result in adverse physical consequences in the real-world such as disruption of operation or deviation from flight paths. De-authentication attacks result in Denial-of-Service for drone operation. Thus, the capacity to detect and mitigate de-authentication attacks is important for the safety and security of UAVs. Therefore, in this paper we propose an intrusion detection system that utilizes diverse network-based features to detect de-authentication attacks using machine learning. The system is trained using super-vised learning on several features extracted from network packets captured from drone operations, to identify attack traffic from normal operational traffic. The experiments are performed on a realistic reference de-authentication attack dataset consisting of a total of 21096 normal and DoS attack instances. The results of our experiments show that Simple Logistic, Random Forest, Random Tree, SVM, and Adaboost machine learning classifiers were able to detect the attacks from the test sets with accuracy ranging from 92.37% to 95.44%. These results underscore the effectiveness of detecting wireless de-authentication attacks using machine learning classifiers trained from features extracted from network packets captured during drone operations.