基于梯度提升决策树模型的P2P网贷借款人信用风险评测研究  

Research on Credit Risk Assessment of P2P Net Loan Borrowers Based on Gradient Boosting Decision Tree Model

作　　者：谭中明[1] 谢坤[1] 彭耀鹏 TAN Zhong-ming;XIE Kun;PENG Yao-peng(School of Finance and Economics,Jiangsu University,Zhenjiang 212003;School of Computer Science and Techrwlogy,Zhejiang University,Hangzhou 310058)

机构地区：[1]江苏大学财经学院,江苏镇江212013 [2]浙江大学计算机科学与技术学院,杭州310058

出　　处：《软科学》2018年第12期136-140,共5页Soft Science

基　　金：国家社会科学基金项目(16BGL049)

摘　　要：基于借款人决策行为视角,通过Logistic条件回归方程式,筛选出对主体行为决策具有显著影响的特征变量,依此构建基于梯度提升决策树(GBDT)的P2P网贷借款人信用风险评测模型。模型精度和稳定性检验结果表明,建立在集成学习基础上的GBDT模型能较好拟合网络信用环境下借款人信用风险评测,并能对样本借款人的决策行为做出准确预判。Based on the borrower's decision-mak!ng behavior,this paper first uses the Logistic condition regression method toselect the characteristic variables which have significant influence on the subject's behavior decision-making.On this ba- sis,a credit risk evaluation model of P2P loan borrower based on gradient boosting decision tree (GBDT)is constructed and further testes the accuracy and stability of the model.Results show that the GBDT model based on the integrated learning can fit the credit risk of the borrower under the network credit environment,and can make accurate judgment on the deci- sion:making behavior of the sample borrowers.

关 键 词：P2P网贷 信用风险评测 梯度提升决策树(GBDT)模型 

分 类 号：F830[经济管理—金融学]