Predicting healthcare costs using GAs

Predicting prospective healthcare costs is of increasing importance. Genetic search is used to discover attribute sets and associated posterior probability classifiers that predict the top 0.5% most costly individuals in year N + 1 based on previous medical conditions and costs in year N. The predictive performance of single-variable classifiers (cost-drivers), found using statistical measures familiar from datamining, as well as Naive Bayesian analysis, are compared and contrasted with that of classifiers found using genetic search. Comparison is also made with two well known benchmarks from the healthcare literature. Categories and Subject Descriptors I.5.1 [Computing Methodologies]: Pattern RecognitionDesign Methodology[Classifier design and evaluation, feature evaluation and selection] General Terms Performance, Measurement Keywords Data mining, healthcare, prediction, costs, genetic algorithm, classifier, search

• Type: Conference
• Year: 2005
• Where: GECCO
• Authors: Christopher R. Stephens, Henri Waelbroeck, S. Talley
• Keywords: Associated Posterior Probability, GECCO 2005, Genetic Search, Prospective Healthcare Costs