Predicting Facilities that Repetitively Support Hostile Activities using Adversary Modeling and Geographic Profiling

Stottler Henke proposes to develop a software system to predict which facilities in an urban environment are likely to repetitively support activities by hostile actors. This system will draw upon an extensible, heterogeneous geographic information system containing detailed geographic and facility data as well as a knowledge base storing information about past hostile actions and known adversary tactics, structure, and associations. The system will employ a combination of geographic profiling techniques and adversary decision models – which capture the facility preference criteria and constraints of the adversary -- to dynamically generate effective predictions based on the latest information. Using machine learning algorithms, the system will automatically adapt its adversary decision models to account for regional differences or changes in adversary tactics over time. The system will also exploit the structure of the knowledge base ontology to generalize and aggregate adversary knowledge where necessary to mitigate gaps in the data. During the Phase I, Stottler Henke will characterize the data sets required for effective prediction of supporting facilities. We will also develop predictive algorithms as described above as well as learning algorithms to adapt to evolving adversary tactics. Finally, we propose to demonstrate the feasibility of our approach via a limited prototype.