Credit scoring based on tree-enhanced gradient boosting decision trees
作者：
Highlights：
• Two tree-based augmented GBDTs is proposed to improve the performance of credit scoring.
• Tree-based embedding improves the efficiency of augmentation for GBDT.
• The intrinsic interpretability of the proposed methods is studied.
• The marginal contribution of features is investigated by embedding TreeSHAP into tree-enhanced GBDT.
• Tree ensembled framework well balance the performance, efficiency, and interpretability of credit scoring.
摘要
•Two tree-based augmented GBDTs is proposed to improve the performance of credit scoring.•Tree-based embedding improves the efficiency of augmentation for GBDT.•The intrinsic interpretability of the proposed methods is studied.•The marginal contribution of features is investigated by embedding TreeSHAP into tree-enhanced GBDT.•Tree ensembled framework well balance the performance, efficiency, and interpretability of credit scoring.
论文关键词：Credit scoring,Gradient boosting decision trees,Feature augmentation,Interpretability
论文评审过程：Received 14 January 2021, Revised 21 September 2021, Accepted 2 October 2021, Available online 20 October 2021, Version of Record 26 October 2021.
论文官网地址：https://doi.org/10.1016/j.eswa.2021.116034
