Disease outbreak detection and tracking for biosurveillance: a data fusion approach

Abstract

In this paper we present an application that utilizes a novel two-level fusion architecture to detect and track disease outbreaks across public health system databases. In the first fusion level, collected data is used to detect and track indicative bio-events using latent semantic analysis and unsupervised clustering. In the second fusion level, clusters produced via the first are used to feed dynamic Bayesian networks which assess outbreak type and state. We train and test our system using data from a 200K+ free-text emergency department (ED) chief complaint record set.