基于人工智能的Lasso-GBDT信用卡风险评级方法  

A Credit Card Risk Assessment Method Based on Lasso-GBDT Model

作　　者：司孟慧 郭威 陈传龙 SI Meng-hui;GUO Wei;CHEN Chuan-long

机构地区：[1]中央党校(国家行政学院) [2]中央党校(国家行政学院)经济学教研部

出　　处：《农村金融研究》2022年第5期28-38,共11页Rural Finance Research

基　　金：国家社会科学基金重大项目“重大国际和地区金融危机发生机理、预警机制和防范政策研究”(编号:18VFH004)的研究成果。

摘　　要：在信贷信息不对称现状下,构建可靠的个人信用评分模型等信用评级方法评估贷款人的信用违约风险水平具有重要的现实意义。论文将具有互补性的Lasso-GBDT模型组合引入个人信用评级,发现Lasso-GBDT组合模型能够准确地筛选出重要变量;通过对商业银行个人信用评级进行实证分析发现,相较于Lasso-RF模型,LassoGBDT模型更能在抓住信用风险关键因素的基础上准确预测信用卡违约状况。Under the current situation of asymmetric credit information,it is of great practical significance to construct a reliable personal credit scoring model to evaluate the credit default risk level of lenders.In this paper,the complementary Lasso-GBDT model is introduced into personal credit rating,and it is found that Lasso-GBDT model can accurately screen out important variables.Through the empirical analysis of personal credit rating of commercial banks,it is found that compared with Lasso-RF model,Lasso-GBDT model can accurately predict credit card default situation based on the key factors of credit risk.

关 键 词：人工智能 信用评级 Lasso-GBDT组合模型 

分 类 号：F832.2[经济管理—金融学]