Mining Bayesian Networks to Forecast Adverse Outcomes Related to Acute Coronary Syndrome

Andrew J. Novobilski, Francis M.Fesmire, and David Sonnemaker

One fascinating aspect of tool building for datamining is the application of a generalized datamining tool to a specific domain. Often times, this process results in a cross disciplinary analysis of both the datamining technique and the application of the results to the domain itself. This process of cross-disciplinary analysis often leads not only to improvements of the tool, but more importantly, to a better understanding of the underlying domain model for the domain experts involved. This paper presents the results of applying a datamining tool for identifying a Bayesian Network to represent a dataset of triage information taken from patients arriving at the emergency room with symptoms of Acute Coronary Syndrome. Specifically, a domain expert generated Bayesian Network and a mined Bayesian Network, both trained using the triage dataset, are compared for their accuracy in forecasting 30-day adverse outcomes for the patients represented in the dataset. The comparison, done using ROC curves, shows that the mined Bayesian Networked slightly outperformed the domain expert generated network. The results are discussed and direction for future work based on the complexity of the mined network versus the expert’s network are presented..

This page is copyrighted by AAAI. All rights reserved. Your use of this site constitutes acceptance of all of AAAI's terms and conditions and privacy policy.