Compressed partial least squares regression: A supervised method for multi-label data
Zongjie Ma, Huawen Liu, Zheng Zheng, Jianmin Zhao, Xiaodan Xu · 2014
Multi-label classification allows an instance to be associated with multiple labels. Compared with other classification tasks, multi-label classification also suffers from the problem of high data dimension. However, the existing dimensionality reduction (DR) methods are not very appropriate for multi-label data. In this paper, we proposed a supervised DR method, named the compressed partial least squares regression for multi-label data (CRMD). First, CRMD aims at reducing the dimensionality of instance space and label space simultaneously, and then establishing the regression model between the two spaces for prediction. Specially, we apply 2-norm penalization on partial least squares to overcome the high dimensionality. The experimental results on six standard public datasets validate the performance of our approach.