A Multi-Label Classification Algorithm Based on Self-paced Learning and LDA

To address the problem that existing multi-label learning algorithms treat all samples equally during model training and ignore inter-sample variability, this paper proposes a multi-label classification algorithm based on self-paced learning and LDA (SPLDA for short). SPLDA first constructs an adaptive self-paced function based on the label prediction loss values in multi-label data, and introduces a self-paced learning framework. Secondly, the linear discriminant analysis regularization term in the low-dimensional space of the samples is constructed by combining the principles of minimizing the within-class distance and maximizing the between-class distance in this framework. Next, a multi-label classification model is trained using alternating optimization methods. Finally, the experimental results on 5 real datasets and 3 evaluation metrics show that SPLDA outperforms the existing multi-label classification algorithms.