A Machine Learning Approach for Detecting Unauthorized Drone Operators

Governmental authorities worldwide have always been striving to impose further restrictions on the use of drones for safety reasons. On the other hand, hijackers always try to find a way to control the UAV to steal it, damage it or even assign tasks that are not meant to be operated. Classification of the UAV operators from the driving behavior is crucial to prevent malicious attacks and hijacking. This paper proposes a short-time prediction of unauthorized pilot behavior. A comprehensive analysis is conducted to find the optimal machine learning approach to classify the UAV operator in terms of accuracy, sensitivity, and prediction time. The utilized dataset consists of recorded flying sessions of 20 different pilots based on four features, thrust, yaw, pitch, and roll. To balance the dataset, the Synthetic Minority Over-sampling Technique (SMOTE) is utilized. A time-series forecasting approach is proposed to recognize repetitive patterns of pilot behavior over short time intervals. Finally, the proposed decision tree (DT) algorithm achieved the highest accuracy with 95% at a prediction time of 15 ms. The results outperformed state-of-the-art solutions for unauthorized pilot detection.