Multi-grained and multi-layered gradient boosting decision tree for credit scoring

Credit scoring is an important process for banks and financial institutions to manage credit risk. Tree-based ensemble algorithms have made promising progress in credit scoring. However, tree-based ensemble algorithms lack representation learning, making them cannot well express the potential distribution of loan data. In this study, we propose a multi-grained and multi-layered gradient boosting decision tree (GBDT) for credit scoring. Multi-layered GBDT considers the advantages of the explicit learning process of tree-based model and the representation learning ability to discriminate good/bad applicants; multi-grained scanning augments original credit features while enhancing the representation learning ability of multi-layered GBDT. The experimental results on 6 credit scoring datasets show that the hierarchical structure can effectively reduce the intra-class distance and increase the inter-class distance of the credit scoring dataset. In addition, Multi-grained feature augmentation effectively increases the diversity of prediction and further improves the performance of credit scoring, providing more precise credit scoring results.