Study of Hellinger Distance as a splitting metric for Random Forests in balanced and imbalanced classification datasets

作者：

Highlights：

• Hellinger Distance (HD) is a robust splitting metric for Random Forests (RF).

• HD is statistically better for imbalanced datasets with respect to AUC.

• HD is statistically better for two-class datasets with respect to Brier score.

• The combination of HD and Gini improves the robustness of RF.

摘要

•Hellinger Distance (HD) is a robust splitting metric for Random Forests (RF).•HD is statistically better for imbalanced datasets with respect to AUC.•HD is statistically better for two-class datasets with respect to Brier score.•The combination of HD and Gini improves the robustness of RF.

论文关键词：Hellinger Distance,Imbalanced problems,Random Forests

论文评审过程：Received 12 July 2019, Revised 31 October 2019, Accepted 30 January 2020, Available online 31 January 2020, Version of Record 7 February 2020.

论文官网地址：https://doi.org/10.1016/j.eswa.2020.113264