A Feature Selection Method Using Hierarchical Clustering

Abstract

Feature selection refers to a problem to select a subset of features which are most optimal for intended tasks. As one of well-known feature selection methods, clustering features into several groups and picking one feature from each group have been used for unsupervised feature selection. Since the purpose of clustering in feature selection is to select a feature from each group, the quality of the feature to be selected should be considered in the clustering process. In this paper, we propose a feature selection method using hierarchical clustering. A new similarity measure between two feature groups is defined by directly using the representative feature in each group. Experimental results show that our method can select good features even for supervised learning.