Course-of-Action Forecasting

Course of Action (COA) forecasting is the capability to integrate temporal and spatial data from multiple intelligence sources, evaluate this data for historical patterns to identify patterns in order to predict future actions at specified spatial locations. In Phase I, we developed a COA Decision Agent based on our learning agent and geospatial similarity approach on an information extraction platform. Our system extracts entities (people, places) and spatiotemporal patterns (events, relationships) from the unstructured intelligence data, categorizes the extracted patterns using taxonomy model, discovers and manages COA facts using belief nets, and enables the intelligence analyst visualize and explore the extracted facts. Our research has shown the feasibility of integrating of multi-source intelligence data, extracting patterns of interest with information extraction tools, generalizing facts on extracted patterns using belief networks based on plan recognition and spatiotemporal knowledge, using this belief system to provide COA forecasting, and visualization of extracted facts in a GIS environment. In Phase II, we propose to develop the full-scale prototype implementation of the COA forecasting system based on our Phase I research and Army intelligence operations process modeling, evaluate its performance with more extensive real-world data, refine the fundamental algorithms based on the evaluation, and transition the research to Phase III commercialization.