论文部分内容阅读
客户信用评估是银行等金融企业日常经营活动中的重要组成部分。一般违约样本在客户总体中只占少数,而能按时还款客户样本占多数,这就是客户信用评估中常见的类别不平衡问题。目前,用于客户信用评估的方法尚不能有效解决少数类样本稀缺带来的类别不平衡。本研究引入迁移学习技术整合系统内外部信息,以解决少数类样本稀缺带来的类别不平衡问题。为了提高对来自系统外部少数类样本信息的使用效率,构建了一种新的迁移学习模型:以基于集成技术的迁移装袋模型为基础,使用两阶段抽样和数据分组处理技术分别对其基模型生成和集成策略进行改进。运用重庆某商业银行信用卡客户数据进行的实证研究结果表明:与目前客户信用评估的常用方法相比,新模型能更好地处理绝对稀缺条件下类别不平衡对客户信用评估的影响,特别对占少数的违约客户有更好的预测精度。
Customer credit rating is an important part of the daily business activities of financial companies such as banks. The sample of general default accounts for only a few of the total customers, and can repay customers on time. The sample is the majority, which is a common category imbalance in customer credit evaluation. At present, the methods used for customer credit assessment can not effectively solve the category imbalance brought by the scarcity of a few sample types. This study introduces migration learning technology to integrate the internal and external information of the system to solve the category imbalance brought by the scarcity of a few types of samples. In order to improve the efficiency of using information from a small number of samples outside the system, a new migration learning model was constructed. Based on the migration bagging model based on integrated technology, two-stage sampling and data packet processing Build and integrate strategies for improvement. The empirical research on credit card customer data of a commercial bank in Chongqing shows that the new model can better deal with the impact of the category imbalance on the credit evaluation of customers under absolute scarce conditions than the commonly used methods of customer credit assessment, A few default customers have better prediction accuracy.