Approximating Dependency for Efficient Multi-label Feature Selection

Abstract

Multi-label feature selection is an important task that can be done before applying multi-label classification algorithms because the multi-label classification performance is naturally influenced by input features. To solve this problem, feature selection algorithm considers the dependency of each feature to labels as well as the dependency among features simultaneously. However, feature selection methods suffer from additional computational burden for calculating the dependency among features. In this paper, we propose an efficient feature selection algorithm extending quadratic programming feature selection for multi-label datasets and use the Nyström approximation. Experimental results demonstrated the proposed method reduces the computational cost for performing multi-label feature selection.