首页|小微企业违约特征再探索:基于SHAP解释方法的机器学习模型

小微企业违约特征再探索:基于SHAP解释方法的机器学习模型

扫码查看
原文链接
  • NETL
  • NSTL
  • 万方数据
机器学习方法已经被应用于小微企业贷款审批和监测过程，并且在违约识别方面取得了良好效果，但是机器学习系统决策过程的不可见性导致其在违约特征识别领域未能得到进一步实际应用。基于某银行的小微企业贷款微观数据，在机器学习模型基础上加入SHAP(SHapley Additive exPlanations)解释方法对小微企业的违约特征进行研究比较，研究兼顾了实际情境中判别准确性和指标可解释的要求。研究发现，除传统的贷款信息与企业财务指标外，违约的核心特征中企业年龄、被告案件数量以及客户经理评价"软信息"等非财务指标对于识别小微企业违约具有重要价值。本文从可解释性的角度探讨机器学习方法在小微企业违约特征识别的应用，创新性地引入SHAP解释方法研究评级中的重要指标，同时所挖掘的关键指标对贷款业务开展具有指导意义。
Re-exploration of Small and Micro Enterprises'Default Characteristics Based on Machine Learning Models with SHAP
Machine learning methods have been applied to the small and micro enterprises'loan approval and monitoring process,and have achieved good results in default identification.Considering the higher recognition accuracy of machine learning methods,its use of indicator information should be better than traditional models.Therefore,it hopes to dig out the important factors in the judgment of default from the perspective of machine learning in this paper.SHAP is a machine learning interpretation method based on the Shapley value of game theory,which can identify the importance of indicators in the model from the perspective of results.Based on the small and micro enterprise loan account of a bank,SHAP(SHapley Additive exPlanations)is added to machine learning models to find important default characteristics of small and micro enterprises.It is found that,in addition to traditional loan information and corporate financial indicators,non-financial indicators such as the age of the company,the number of law cases,and the"soft information"evaluated by the customer manager play significant role in identifying defaults of small and micro enterprises.From the perspective of interpret-ability,the application of machine learning methods is discussed in the identification of default characteristics of small and micro enterprises,and innovatively the SHAP interpretation method is introduced to study important indicators in rating.At the same time,the key indicators mined have guiding significance for the development of loan business.

small and micro enterprisesdefault characteristicsnon-financial informationSHAPmachine learning

雷欣南、林乐凡、肖斌卿、俞红海

展开 >

南京大学工程管理学院,江苏 南京 210093

小微企业 违约特征 非财务信息 SHAP解释方法 机器学习

国家自然科学基金项目国家社会科学基金重大项目国家自然科学基金面上项目国家自然科学基金面上项目国家自然科学基金重点国际合作项目

7234202419ZDA105720711027167108371720107001

2024

10.16381/j.cnki.issn1003-207x.2021.0027

中国管理科学
中国优选法统筹法与经济数学研究会 中科院科技政策与管理科学研究所

中国管理科学

CSTPCDCSSCICHSSCD北大核心
影响因子：1.938
ISSN：1003-207X
年,卷(期)：2024.32(5)