A multi-label classification algorithm based on Partial Least Squares regression
Qiande Ren, Farong Zhong · 2012
In multi-label learning, an instance may be associated with a set of labels, and Multi-Label Classification (MLC) algorithm aims at outputting a label set for each unseen instance. In this paper, a MLC algorithm named ML-PLS is proposed, which is based on Partial Least Squares (PLS) regression. In detail, as PLS can handle the relations between the matrices of independent variables and dependent variables through a multivariate linear model, when PLS is directly used for MLC, the matrix of dependent variables is set to include the information of the label memberships and the labels of dependent variables can then be predicted through the multivariate linear model. Experiments on real-world multi-label data sets show that ML-PLS is significantly competitive to other MLC algorithms.