Predictive Analysis System: a case study of AI techniques for counternarcotics

Abstract

The Predictive Analysis System (PANS) uses knowledge of narco-trafficking behaviors to help analysts fuse all-source data into coherent pictures of activity from which predictions of future events can be made automatically. The system uses a form of model-based reasoning, plan recognition, to match reports of actual activities to expected activities. The model incorporates several sets of domain constraints and a constraint propagation algorithm is used to project known data points into the future (i.e., predict future events). The system can track many possibilities concurrently, and also allows analysts to hypothesize activity and observe the possible effect of the hypotheses on future activities. It makes use of recent results in knowledge representation, plan recognition, and machine learning to capture analysts' expertise without suffering from the brittleness of rule-based expert systems.<>