Preemptive Actions with Dynamic Anticipatory Targeting (PREDATAR)

A target type prediction system dubbed PREDATAR has been designed under a Phase I effort, to provide evolving probabilistic identifications of ground targets nominated as Time Sensitive Targets (TSTs), well in advance of their positive identification (PID). For each TST, a probabilistic identification vector of potential target types is successively refined via a five-step ‘pipelined’ process: 1) expert rules provide a ‘quick expert guess’ about the TST target type; 2) a target ontology is used to determine which target types are feasible; 3) credibility checks of target type versus reporting ISR platform are performed to eliminate obvious errors; 4) learning about target types is leveraged to adapt to new operational situations; and 5) the most likely TST target types are vetted using encapsulated expert knowledge. The viability of the proposed approach was studied in the context of a post-high intensity scenario. Implementation of a PREDATAR prototype has been initiated, based on an in-house Bayesian Belief Network engine. Under Phase II, implementation and validation will be completed, by bringing in additional tools and technologies related to ontology development, decision tree learning, and rule-based inferencing. Future plans include transitioning PREDATAR to existing and ongoing DoD efforts, including ADOCS, TBMCS, GCCS, and DCGS.