An enhanced random forest with canonical partial least squares for classification

Abstract

Recently, several variants of random forest have been derived for the classification problems, among which the rotation forest is an important type to improve the model’s accuracy. In this article, we proposed a simple and effective variation of rotation forest, which the canonical partial least squares algorithm is employed to rotate the variable space of tree and then all the trees are combined being a “forest.” Results of an experiment on a sample of 20 benchmark datasets show our method has better prediction performance comparing with random forest and rotation forest.

Keywords:

• Canonical partial least squares
• random forest
• classification
• feature rotation

Acknowledgment

The authors thank the editor and two reviewers for constructive suggestions.

Additional information

Funding

This work was supported in part by National Social Science Foundation of China (17BTJ019), the National Natural Science Foundation of China (Grant Nos. 11801105 and 11561010).