基于LightGBM与SHAP的信贷违约预测方法研究  

Research on credit default prediction method based on LightGBM and SHAP

作　　者：戴峥琪 雷亿辉[1] 彭晨 夏广萍 DAI Zhengqi;LEI Yihui;PENG Chen;XIA Guangping(School of Mathematics and Statistics,Jishou University,Jishou 416000,China;School of Computer Science and Engineering,Jishou University,Jishou 416000,China)

机构地区：[1]吉首大学数学与统计学院,湖南吉首416000 [2]吉首大学计算机科学与工程学院,湖南吉首416000

出　　处：《邵阳学院学报（自然科学版）》2024年第1期84-91,共8页Journal of Shaoyang University：Natural Science Edition

基　　金：国家自然科学基金(62006095);湖南省自然科学基金(2021JJ40441,2023JJ30485);吉首大学科研基金资助项目(Jdy22010)。

摘　　要：机器学习方法在信贷领域取得了较好的成果,但由于缺乏可解释性,应用受到限制,为增加其可信度和透明度,克服“黑盒”模型缺乏可解释性的缺陷,基于LightGBM算法建立信贷违约预测模型,并设计SHAP算法对模型的结果进行解释。结果表明,模型性能更好,预测精度更高,其精度高达88.61%;SHAP算法解释结果表明“信用组合的分类”“要支付的剩余债务”“每月EMI付款”等因素对信贷决策有着重要影响。Machine learning methods have shown promising results in the credit domain;however,their application is constrained by a lack of interpretability.To enhance credibility and transparency,and overcome the opacity inherent in“black box”models,a credit default prediction model based on the LightGBM algorithm is established.Additionally,the SHAP algorithm is employed to elucidate the model s outcomes.The findings indicate superior performance of the proposed model,achieving an impressive prediction accuracy of 88.61%.Furthermore,SHAP algorithm interpretations reveal the significance of factors,such as“Credit-Mix”“Outstanding_Debt”and“Total_EMI_per_month”in influencing credit decisions.

关 键 词：信贷风险 LightGBM算法 SHAP算法 可解释性 

分 类 号：F832.4[经济管理—金融学]