[1]郭畅.基于不平衡数据的个人信贷违约测度探索[J].集美大学学报(自然版),2021,26(1):89-96.
 GUO Chang.Personal Credit Default Measurement Research Based on the Imbalanced Data[J].Journal of Jimei University,2021,26(1):89-96.
点击复制

基于不平衡数据的个人信贷违约测度探索()
分享到:

《集美大学学报(自然版)》[ISSN:1007-7405/CN:35-1186/N]

卷:
第26卷
期数:
2021年第1期
页码:
89-96
栏目:
数理科学与信息工程
出版日期:
2021-01-28

文章信息/Info

Title:
Personal Credit Default Measurement Research Based on the Imbalanced Data
作者:
郭畅
(安徽大学经济学院,安徽 合肥 230601)
Author(s):
GUO Chang
(School of Economics,Anhui University,Hefei 230601,China)
关键词:
信贷风险违约预测类别不平衡集成模型子模型个数
Keywords:
credit riskdefault predictionclass imbalanceensemble modelthe number of sub models
摘要:
针对个人信贷风险评估中存在的类别不平衡问题,为了提升信贷违约客户的识别能力,提出基于欠采样改进的集成模型。该模型从“数据”层面进行批量欠采样处理,从“算法”层面对现有的集成模型进行再次集成。在UCI台湾信用卡信贷数据集上,结合模型整体效果的测度AUC值、精度方面的测度F1值和区分度指标KS值进行评估。结果表明,基于欠采样改进的Batch-US-集成模型总能取得更优结果,其中属Batch-US-Xgboost模型最优,接着对Batch-US-集成模型的子模型个数和模型有效性进行探索,证实了子模型个数并非越多越好的结论。改进后的Batch-US-集成模型能够有效提升信贷风险违约预测的效果。
Abstract:
Aiming at the problem of category imbalance in personal credit risk assessment,in order to improve the identification ability of credit default customers,an improved integration model based on under sampling is proposed.The model is based on under sampling processing from the data level and reintegrating the existing integration model from the algorithm level,and studies the improvement effect of this model.On the UCI Taiwan credit card credit data set,it evaluates the AUC value of the overall effect of the model,the F1 value of the accuracy and the KS value of the differentiation index .The results show that the Batch -US-Ensemble models based on the under sampling process can always achieve better results,and the Batch-US-Xgboost model is the best among all ensemble models.Then,the number of sub models and the validity of Batch-US-ensemble models are explored,which proves that the number of sub models is not the more the better.The improved Batch-US-Ensemble model can effectively improve the effect of credit risk default prediction.
更新日期/Last Update: 2021-03-25