Feature selection in multi-label classification through MLQPFS

Abstract

In multi-label classification, each data instance is associated with a set of labels. Feature selection is one of the most significant challenges in multi-label classification. Irrelevant and dependent features can mislead the learning phase of multi-label classification. Therefore it is important to select the effective features. A large and growing body of literature has investigated multi-label feature selection problem. Most of these studies used incremental approach. In this paper a new algorithm has been proposed called MLQPFS which selects subset of features in such a way that redundancy among the selected features will be minimized meanwhile the relevancy between the selected features and class labels will be maximized. MLQPFS applies quadratic programming to optimize feature selection process. In order to evaluate the performance of proposed algorithm, MLQPFS and PMU have been compared in three multi-label data sets. The experimental results showed that MLQPFS has better performance than conventional incremental feature selection methods such as PMU.