An SVM classifier incorporating simultaneous noise reduction and feature selection: illustrative case examples

摘要

A hybrid technique involving symbolization of data to remove noise and use of conditional entropy minima to extract relevant and non-redundant features is proposed in conjunction with support vector machines to obtain more robust classification algorithm. The technique tested on three data sets shows improvements in classification efficiencies. (C) 2004 Pattern Recognition Society. Published by Elsevier Ltd. All rights reserved.