Quantifying the robustness of a Bayesian Belief Network in the context of Unmanned Aerial System threat prediction

Abstract

Unmanned Aerial Systems (UASs), or drones, are becoming increasingly available to the general public. Because of this, organizations active in safety and security, such as the Dutch Armed Forces and the Dutch National Police, need to be prepared for possible UAS accidents and attacks. To that end, it is vital that the nature of the possible threat that UAS may pose is detected in a timely manner. A method that can be employed for this problem is a Bayesian Belief Network (BBN). Given the observations made of a UAS and its surroundings, a BBN is capable to determine the most likely type of threat posed by the UAS. Generally, the probabilities that are required as input for this method can be estimated from historical data if enough data are available. However, since only a small amount of data about drone incidents has been collected, expert opinion is used. This introduces uncertainty in the BBN as opinions of experts are subjective. This paper presents a means to construct a BBN for UAS threat prediction when no empirical data are available and determine the robustness of the output. The analysis is restricted specifically to NATO Class I drones (less than 150 kg) in law enforcement operations.