Gradient and Newton boosting for classification and regression

作者：

Highlights：

• Present gradient, Newton, and hybrid gradient-Newton boosting in a unified framework.

• Show that Newton boosting achieves significantly higher predictive accuracy.

• Reason for higher predictive accuracy is not faster convergence.

• Introduce novel interpretable tuning parameter which is important for predictive accuracy.

摘要

•Present gradient, Newton, and hybrid gradient-Newton boosting in a unified framework.•Show that Newton boosting achieves significantly higher predictive accuracy.•Reason for higher predictive accuracy is not faster convergence.•Introduce novel interpretable tuning parameter which is important for predictive accuracy.

论文关键词：Boosting,Supervised learning,Ensembles,Trees

论文评审过程：Received 23 July 2019, Revised 30 September 2020, Accepted 30 September 2020, Available online 20 October 2020, Version of Record 10 February 2021.

论文官网地址：https://doi.org/10.1016/j.eswa.2020.114080