基于MC3投票法和机器学习的信用风险评估  

Credit risk assessment based on the MC~3 voting method and machine learning

作　　者：邝雄 张成祖 张婷婷 KUANG Xiong;ZHANG Chengzu;ZHANG Tingting(International Business School,Hainan University,Haikou 570228,China)

机构地区：[1]海南大学国际商学院,海南海口570228

出　　处：《海南大学学报（人文社会科学版）》2025年第1期97-106,共10页Journal of Hainan University （Humanities & Social Sciences）

基　　金：国家自然科学基金项目(71963011,72163004);海南省基础与应用基础研究计划(自然科学领域)高层次人才项目(2019RC065)。

摘　　要：信用风险评估是金融风险管理的重要问题,为提高信用风险评估的有效性,基于马尔科夫链蒙特卡罗模型综合方法提出模型投票方法,这种方法可以不需要进行指标剔除,减少了特征选择过程中的信息丢失,同时可以更为谨慎地评估信用风险。结合MC3投票法和机器学习方法,构建了信用风险评估模型。在此基础上,借助国泰安数据库上市制造业企业的财务指标数据,对构建的信用风险评估模型与其他模型的预测性能进行了比较分析。检验结果表明:相对于一次剔除法和逐步剔除法,MC3投票法降低了银行由于信用风险评估模型的一类错误而造成的损失,从而提高了信用风险评估模型的性能。The credit risk assessment is an important issue of financial risk management.In order to improve the effectiveness of credit risk assessment,this paper first proposes a model voting method based on the Markov Chain Monte Carlo Model Composition.This method can eliminate the need for indicator elimination,reduce the loss of information in the process of feature selection,and assess the credit risk more carefully.Then,combined with the MC~3 voting method and machine learning method,a credit risk assessment model is constructed.On this basis,it compares the prediction performances of this model and other models with the help of the financial index data of listed manufacturing enterprises in the China Stock Market Accounting Research Database.The test results show that the MC~3 voting method,compared with the one-time elimination method and the step-by-step elimination method,reduces the losses caused by Class 1 error in the credit risk assessment model,and thus improves the performance of the credit risk assessment model.

关 键 词：信用风险评估模型 机器学习 MC3投票法 

分 类 号：F830.5[经济管理—金融学]