Feature selection via genetic optimization

Abstract

In this paper we present a novel Genetic Algorithm (GA) for feature selection in machine learning problems. We introduce a novel genetic operator which fixes the number of selected features. This operator, we will refer to it as m-features operator, reduces the size of the search space and improves the GA performance and convergence. Simulations on synthetic and real problems have shown very good performance of the m-features operator, improving the performance of other existing approaches over the feature selection problem. © Springer-Verlag Berlin Heidelberg 2002.