Modeling and Prediction of Asymmetric Threat Learning Processes

The overall goal of this research is to develop an automated tool for enemy course of action prediction that accounts for enemy learning. This tool will be grounded in formal techniques that provide testable results. In Phase I, CCRI will focus on the development of the formal, mathematical techniques that will enable us to understand and predict an asymmetric enemy’s attack patterns and learning behavior. To accomplish this goal, CCRI will: 1. Refine their existing discrete choice models for asymmetric warfare event prediction to enable the detection of adaptive behavior by our opponents; 2. Revise their existing search methods to enable the rapid discovery of important features in the enemy’s attack planning and learning; 3. Develop an ontology that links the features used by asymmetric warfare opponents in their selection of sites and times of attack, as well as, their learning; 4. Develop techniques that enable the rapid modification of the discrete choice models to predict enemy behavior in the presence of adaptive behavior by the enemy; 5.Incorporate dynamic features in their predictive models and improve their techniques for mining these features to enable the detection of significant change and importance to the behavior of our enemy; and 6. Refine their existing discrete choice models to actually predict and anticipate the future actions of the enemy in the presence of learning behaviors. BENEFITS: Under Phase I, CCRI will develop the techniques for enemy course of action prediction that account for enemy learning. As noted in the task descriptions in Section 3.1, each task will produce significant results toward this Phase I goal. Specifically CCRI will produce: 1. Predictive modeling techniques for asymmetric warfare incidents that explicitly model temporal components; 2. An ontology for asymmetric warfare attack prediction that accounts for learning; 3. Measures of feature importance for asymmetric warfare behavior; 4. Techniques to detect changes in asymmetric warfare attack behavior; 5. Methods that semantically link features relevant to asymmetric warfare attacks, so that measured changes in the importance of a feature are understood in the context of the relative importance of other features; 6. Techniques that integrate all of the above results to enemy courses of action in the presence of learning. The work CCRI has proposed for this SBIR would have applicability to the development of new analytical tools for law enforcement. CCRI is currently engaged in discussions with DaPro Systems, Inc., a leading provider of RMS to law enforcement agencies in Virginia, about creating analytical tools for law enforcement. These tools would be compatible with the Records Management System (RMS) provided by DaPro Systems and allow crime analysts to identify patterns of criminal behavior and allocate resources to better address these behaviors.