Date of Award

3-2004

Document Type

Thesis

Degree Name

Master of Science

Department

Department of Operational Sciences

First Advisor

Kenneth W. Bauer, PhD

Abstract

This thesis continues the research begun by Storm, Bauer and Oxley in 2003 into the fusion of classifiers. It examines the fusion of up to three correlated classifiers using three different fusion techniques. The overall objective was to determine the optimal ensemble of classifiers to maximize the expected classification accuracy. The ISOC fusion method (Haspert, 2000), the ROC Within fusion method (Oxley and Bauer, 2002) and a Probabilistic Neural Network were the three fusion techniques employed in these set of experiments. Performance of the classifiers and the fusion methods is measured via ROC curves. Two possible configurations of feature correlations were examined. The expected true positive value relative to a prior distribution of correlation levels for each configuration was then used to compare the classifiers and the fused classifiers performance and thereby allowing for the selection of an optimal ensemble.

AFIT Designator

AFIT-GOR-ENS-04-03

DTIC Accession Number

ADA422890

Share

COinS