Weighting based approach for semi-supervised feature selection

Abstract

Semi-supervised feature selection has become more important as the number of features has increased in partially labeled data sets. In this paper we present a feature weighting-based model to address this problem. Our proposal is based on a semi-supervised clustering paradigm that can rank features according to their relevance from high-dimensional data. We propose an adaptation of the constrained KMeans algorithm to semi-supervised feature selection by an embedded approach. Experiments are provided on several known data sets for validating our proposal. The results are promising and competitive with several representative methods.