Extended adaptive Lasso for multi-class and multi-label feature selection

Abstract Feature selection is a basic step and important task in applications of pattern recognition and machine learning. In this paper, we propose a new Extended Adaptive Least absolute shrinkage and selection operator (EALasso) feature selection method for multi-class and multi-label learning problems. It preserves the oracle properties of identifying the correct subset model and having the optimal estimation accuracy. An iterative optimization algorithm, with theoretical proof of convergence, is proposed to solve the optimization problem. Experimental results on several real multi-class and multi-label data sets show that the proposed EALasso method outperforms many state-of-the-art feature selection methods for multi-class single-label classification as well as multi-class multi-label classification.