Detection of rare events with uncertain outcomes

Abstract

Choquet Expected Utility framework for decision making under uncertainty is compared with the Expected Utility framework in a scenario involving rare events with potentially catastrophic consequences. The scenario involves detection of suspected criminal activity in a surveillance system. The results of simulation studies show the advantages of using the Choquet framework to model pessimistic attitude of decision maker. The Choquet approach detects threats earlier and with higher probability of detection when compared to Expected Utility approach. The paper outlines the relevant decision theory, provides details of simulation design, and experimental results.