个人信贷违约预测模型的研究  

Prediction models for personal credit default

作　　者：周翔[1] 张文宇[1] 江业峰[1] ZHOU Xiang;ZHANG Wenyu;JIANG Yefeng(School of Computer Science and software Engineer,University of Science and Technology Liaoning,Anshan 114051,Chaina)

机构地区：[1]辽宁科技大学计算机与软件工程学院,辽宁鞍山114051

出　　处：《辽宁科技大学学报》2020年第3期223-230,共8页Journal of University of Science and Technology Liaoning

基　　金：辽宁省自然科学基金(20180551011)。

摘　　要：采用信息化手段防控信贷违约风险对保障信贷产业健康发展具有重大的现实意义。传统信贷违约预测模型风险防控能力有限,且在应用有效性评价指标缺乏统一标准。为此,分别基于支持向量机,贝叶斯及随机森林方法建立了信贷违约预测模型,并提出了由准确率、AUC(Area Under ROC Curve)及漏警率所组成的模型性能综合评价指标,同时分别应用信贷信息原始数据和经特征提取后的数据,针对以上三个预测模型性能进行的对比实验表明,由于信贷违约数据具有小基数特征,常规的对初始数据进行特征提取的方法会丧失数据间的高维关联,将严重影响模型的预测效果;同时,基于随机森林的违约预测模型,较其它两个模型表现出更为优异的性能,准确率达到91.2%,综合评价指标达到81.7%,更适用于信贷违约预测领域。It is of great significance to ensure healthy development of credit industry by reducing credit default risk with information technologies.The traditional prediction models has limited capability in controlling credit default risks,and needs standard evaluation indexes.Therefore,three credit default prediction models were built based on support vector machine,Bayes and random forest methods,respectively,and comprehensive evaluation indexes including accuracy rate,area under ROC curve(AUC),and missing alarm rate were proposed.Performance of the three models was compared by using original credit data and feature-extracted data,respectively.The results show that the conventional feature-extraction method tends to lose high dimensional association of the initial credit default data with characteristics of small base number,so that the prediction effects of the model are seriously affected.Meanwhile,compared with the other two models,the model based on the random forest method shows better performance with an accuracy of 91%and a comprehensive evaluation index of 81.7%,being more practical for prediction territory of credit defaults.

关 键 词：信贷违约预测 贝叶斯 支持向量机 随机森林 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程] F830.5[自动化与计算机技术—控制科学与工程]