An efficient algorithm for feature selection with feature correlation

Abstract

Feature selection is an important component of many machine learning applications. In this paper, we propose a new robust feature selection method for multi-class multi-label learning. In particular, feature correlation is added into the sparse learning of feature selection so that we can learn the feature correlation and do feature selection simultaneously. An efficient algorithm is introduced with rapid convergence. Our regression based objective makes the feature selection process more efficient. Experiments on benchmark data sets illustrate that the proposed method outperforms many state-of-the-art feature selection methods. © Springer-Verlag 2013.