机器学习视角下商业银行客户信用风险评估研究  

Research on Customer Credit Risk Assessment of Commercial Banks from the Perspective of Machine Learning

作　　者：顾洲一 胡丽娟[1] Gu Zhouyi;Hu Lijuan(Zhejiang Financial College,Hangzhou 310018,Zhejiang,China)

机构地区：[1]浙江金融职业学院,浙江杭州310018

出　　处：《金融发展研究》2022年第1期79-84,共6页Journal Of Financial Development Research

基　　金：教育部人文社会科学规划基金项目“基于大数据的城市信用监测与评价体系研究”(18YJC790117)。

摘　　要：有效把控信贷风险是商业银行稳健运行的关键环节。本文从商业银行客户信贷数据出发,运用非平衡样本处理算法使少数类样本信息得到平衡,并通过机器学习分类器挖掘影响客户违约的重要风险因子,最后构建Logistic模型计算违约概率。研究发现:第一,客户忠诚度是重要因子,忠诚度越高,客户违约概率越低;第二,客户历史信贷数据价值高,是事前风险控制中的重要参考依据;第三,信贷合同特征是影响客户违约的另一重要维度,包括合同期限和合同利率。研究结论可以为银行授信、风险预警和防范违约风险提供理论参考和实践指导。Effectively controlling credit risk is the key link for the steady operation of commercial banks.Based on the customers'credit data of commercial banks,this paper uses an unbalanced sample processing algorithm to balance the information of minority samples,and mines the key risk factors affecting customer default by a machine-learning classifier.Finally,a Logistic Model is constructed to calculate the default probability.It is found that:firstly,customer loyalty is an important fundamental factor;the higher the loyalty,the lower the chance of customer default;secondly,high value of historical customer credit data,which is an important reference basis in ex ante risk control;thirdly,credit contract characteristics are another important dimension affecting customer default,including contract duration and contract interest rate.The findings of the study can provide theoretical references and practical guidance for bank credit granting,risk warning and default risk prevention.

关 键 词：信贷风险 非平衡处理 机器学习 LOGISTIC模型 

分 类 号：F830.5[经济管理—金融学]