Behavior Anomaly Detection

Abstract

Modern warfare demands situational awareness of entities in the environment. To enhance the warfighter’s situational awareness, we developed an algorithm that detects anomalous behavior in the warfare environment. Changes in entities’ behavior can be an indicator that existing prediction models or assumptions must be updated to remain useful for decision-making. Specifically, we introduce a new classification method—sequential sample consensus (SeqSAC)—that identifies anomalous behavior based on a series of observations of an entity. SeqSAC can support a wide variety of models from simple to complex. We first demonstrate the utility of SeqSAC with a simple limited-degree-of-freedom kinematic model of a moving body, and then we demonstrate the ability to incorporate more complex models using the finite-state machine in Advanced Framework for Simulation, Integration and Modeling (AFSIM). Finally, we discuss the ability to extend SeqSAC to identify anomalies in coordinated entity behaviors.