Automated Link Analysis for Data-mining of Distributed Information (ALADDIN II)

Recent terrorist actions, particularly the delivery of anthrax via the US Postal Service, have triggered an urgent need for an early warning system for disease outbreaks due to bioterrorism. In response, we have developed, under a Phase I SBIR effort, a Bayesian Belief Network (BN) approach for Automated Link Analysis for Data-mining of Distributed INformation (ALADDIN) for early outbreak detection. We developed a limited-scope ALADDIN prototype using our in-house BN engine BNET 2000 and the COTS link analysis engine StrategistT to generate alerts, and demonstrated feasibility of the approach via simulated bioterrorist attack events provided by BioWar. For Phase II, we propose full-scope development of ALADDIN. We will augment the probabilistic BN framework with a Sequential Probability Ratio Test (SPRT) to obtain a termination criterion for the evidence accumulation process. We plan to employ Haddon''s Matrix methodology to facilitate building BN models. The prototype will be tested and validated with spatio-temporal data generated from a variety of simulated scenarios using BioWar, and covering a comprehensive set of bio-agents, diseases, and symptoms, thus enhancing the alert generation capability for unknown bio-agents via model mismatch. We also plan to transition our Phase II product to the Air Force's ESSENCE and LEADERS programs.