NLC: A measure based on projections

Abstract

In this paper, we propose a new feature selection criterion. It is based on the projections of data set elements onto each attribute. The main advantages are its speed and simplicity in the evaluation of the attributes. The measure allows features to be sorted in ascending order of importance in the definition of the class. In order to test the relevance of the new feature selection measure, we compare the results induced by several classifiers before and after applying the feature selection algorithms. © Springer-Verlag Berlin Heidelberg 2003.