Multi-label feature selection with missing labels

Pengfei Zhu, Qian Xu, Qinghua Hu, Changqing Zhang, Hong Zhao

Abstract: Highlights•This is the first attempt to conduct feature selection for multi-label classification with missing labels.•An embedded feature selection method is proposed with which feature selection can be conducted during the process of label recovery.•The effective l_2,p-norm regularization is imposed on the feature selection matrix to select the most discriminative features and remove noisy ones at the same time.•Label dependency is incorporated into the model to exploit label correlations.