Selection of classifiers using information-theoretic criteria

Abstract

Combining multiple classifiers have focused mainly on combination methods, but a few studies have investigated on how to select component classifiers from a classifier pool. Performance by the information fusion varies with the component classifiers as well as the combination method. Previous studies focus on diverse classifiers which accurate and make different errors using the overproduce and choose strategy or the measures of diversity. In this paper, methods based on information theory are proposed for selecting component classifiers by considering the relationship among classifiers. These methods are applied to the classifier pool and examine the possible classifier sets. A classifier set is selected as a candidate and evaluated together with the other classifier sets on the recognition of public unconstrained handwritten numerals. © Springer-Verlag Berlin Heidelberg 2005.