On the use of Machine Learning and Evidence Theory to improve collision risk management

This paper presents an element of Artificial Intelligence (AI) system to assist operators with the management traffic in orbit. The is based on a classification collision events using Dempster–Shaffer’s theory Evidence (DSt). DSt proposed capture and quantify epistemic uncertainty estimation Probability Collision (PC). By capturing calculation PC, we mitigate paradoxical Dilution that affects usual definition this quantity. phenomenon gives counterintuitive idea lower amount information available operators, smaller probability collision. lack corresponds nature. Furthermore, when different sources provide contradictory information, level PC can lead false confidence likelihood either undesirable increase number avoidance manoeuvres or equally negative. In order make automatic continuous decision support investigated use Machine Learning techniques identified algorithm choice deliver correct events. We first considered where Belief Plausibility correctness were used as additional criteria. Then tested second method avoids direct computation but retains same added credibility PC. Results suggest be effectively conjunction render analysis automatic.