Assessing supervised machine learning in detecting GPS spoofing attacks on UAV application use-cases

The use of Unmanned Aerial Vehicles (UAVs) is growing rapidly for multiple purposes, e.g., UAVs play a crucial role in aerial surveillance for road-traffic monitoring or emer-gency response, in smart agriculture for precision fertilization and irrigation, and in delivery of goods. However, due to the increasing attention they draw, UAVs become attractive targets for attackers, e.g., the widespread reliance on Global Positioning System (GPS) technology particularly opens the door for GPS spoofing that manipulates the GPS signals. Machine learning (ML) has revolutionized the field of UAVs security providing the means to analyze data towards detecting attacks, such as unauthorized interference in their navigation systems. In this paper, we elaborate on an open data-set containing GPS relative information regarding different UAV flights, either clear or under spoofing attack, and deploy a number of well-known methods of supervised machine learning as a means of detecting the attacks. We derive quantitative results indicating accuracy, precision, recall, f-score, False Alarm Rate (FAR), Miss Detection Rate (MDR), as well as detection time. Our interest focuses on assessing these methods in the light of diverse UAV application use-cases and highlighting tradeoff issues regarding the metrics of accuracy and detection time. We particularly consider the use-cases of road-traffic surveillance, precision agriculture, and delivery of goods. Our study uncovers the importance of mixing and matching the appropriate ML method in the use-case context balancing complexity with time requirements.