A two-stages unsupervised/supervised statistical learning approach for drone behaviour prediction

Drones are prone to abuse due to their low cost and their pool of potential illegal applications that can compromise safety of national infrastructures and facilities. Hence, drone detection and predict its behaviour is crucial to ensure smooth operation of services. In this paper, an unsupervised/supervised statistical learning algorithm for drone behaviour prediction is proposed. The algorithm is based on drone detection data collected from any radar or RF- sensor. The architecture of the approach is comprised of two stages: i) the first stage attempts to study the drone detection data using either unsupervised or supervised learning methods to model low dimensional expert’s features, and ii) in the second stage a real time drone behaviour predictor model is proposed based on the Kolmogorov-Smirnov and Wasserstein distances. Simulation studies using synthetic data obtained from the AirSim simulator are given to provide the evidence-base for future improvements in the field of drone behaviour prediction.