Selecting features and objects for mixed and incomplete data

Abstract

Selecting objects and features before classifying is a very important task, and can lead to big improvements in classifier accuracy and speed. There are many papers about this topic, but few of them consider the simultaneous or combined approach. In this paper, we present a new method for combined object and feature selection for databases with features not purely numeric or non-numeric. The experiments performed show that it attains the best tradeoff between object and feature reduction in 12 of 15 tested databases, without a significant impact in 1-NN accuracy. © 2008 Springer-Verlag Berlin Heidelberg.