基于进化多目标软子空间聚类的商业银行企业客户信用风险识别

Corporate credit risk identification of commercial banks based on evolutionary multi-objective soft subspace clustering

提出了一种进化多目标软子空间聚类(EMOSSC)算法,用于提升商业银行信贷审批过程中企业客户的信用风险识别和管理水平.考虑到信用数据高维、类不平衡的特征,将聚类算法中单一的聚类有效性指标转化为了一个四目标函数,并采用进化算法对该函数进行优化和求解.结果表明,EMOSSC算法不仅在信用风险识别准确率、稳健性以及结果显著性等方面显著优于对比算法,还能通过对指标权重大小的排序,揭示商业银行企业客户信用风险的关键影响因素,为商业银行的信用风险识别和管理提供有益参考.

In order to promote corporate credit risk identification and management in the credit approval process,an evolutionary multi-objective soft subspace clustering(EMOSSC)algorithm with four objective functions is proposed.Considering the credit datasets are high-dimensional and class-imbalanced,a single clustering validity index in clustering algorithm is transformed into a four-objective function,and then the evolutionary algorithm is used to optimize and solve this function.The results demonstrate that the EMOSSC algorithm is not only significantly better than the other methods in terms of identification accuracy and robustness,but also can reveal key factors by sorting the index weight,which provides a series of useful references for credit risk identification and management.