Simultaneous features and objects selection for mixed and incomplete data

Abstract

In this paper a new simultaneous editing and feature selection method for the Most Similar Neighbor classifier is proposed. It is designed for databases with objects described by features no exclusively numeric or categorical. It is based on Testor Theory and the Compact Set Editing method, mixing edited projections until a good accuracy is achieved. Experimental results with several databases show a good performance compared to previous methods and the classifier using the original sample. © Springer-Verlag Berlin Heidelberg 2006.