A cluster based hybrid feature selection approach

Data collection and storage capacities have increased significantly in the past decades. In order to cope with the increasingly complexity of data, feature selection methods have become an omnipresent preprocessing step in data analysis. In this paper we present a hybrid (filter - wrapper) feature selection method tailored for data classification problems. Our hybrid approach is composed of two stages. In the first stage, a filter clusters features to identify and remove redundancy. In the second stage, a wrapper evaluates different feature subsets produced by the filter, determining the one that produces the best classification performance in terms of accuracy. The effectiveness of our method is demonstrated through an empirical evaluation performed on real-world datasets coming from various sources.