CAJ | 학술논문

现实的银行客户信用评估数据常包含大量的缺失值，这在很大程度上影响了信用评估模型的性能。针对已有模型的不足，提出了面向缺失数据的动态分类器集成选择模型DCESM 。该模型充分利用数据集中所包含的已知信息，在训练信用评估模型之前不需要事先对缺失数据进行预处理，从而减少了对数据缺失机制假设以及数据分布模型的依赖。从UCI数据库中选择两个银行信用卡业务信用评估数据集进行实证分析，结果表明，与4种常用的基于插补法的多分类器集成模型以及1种直接面向缺失数据建模的集成模型相比，DCESM模型能够取得更好的客户信用评估性能。
현실적은행객호신용평고수거상포함대량적결실치，저재흔대정도상영향료신용평고모형적성능。침대이유모형적불족，제출료면향결실수거적동태분류기집성선택모형DCESM 。해모형충분이용수거집중소포함적이지신식，재훈련신용평고모형지전불수요사선대결실수거진행예처리，종이감소료대수거결실궤제가설이급수거분포모형적의뢰。종UCI수거고중선택량개은행신용잡업무신용평고수거집진행실증분석，결과표명，여4충상용적기우삽보법적다분류기집성모형이급1충직접면향결실수거건모적집성모형상비，DCESM모형능구취득경호적객호신용평고성능。
The data in the bank customer ’ s credit scoring often include lots of missing values , which affect the modeling performance to a large extent .To overcome the deficiencies of existing models , this paper proposes a dynamic classifier ensemble selection model for missing values ( DCESM) .The model can make full use of the information included in the dataset and does not need to pre-process the missing values before training the model , which decreases the dependence on the hypothesis for data missing mechanism and distribution model . Two credit scoring datasets on bank credit card business from UCI database were selected for our empirical a -nalysis .The results show that the DCESM model outperformed four imputation-based multi-classifiers ensemble models and one ensemble model for missing values .