摘要
特征选择是信用评级建模的重要环节,合理的特征选择能够简化模型结构和提升分类效果.借鉴w-L1SVM模型的加权思想,借助Logistic-Group-Lasso模型筛选组变量的优势,提出处理信用评级特征选择问题的加权Logistic-GroupLasso(w-LGL)模型,该模型在选择变量时更加关注误判成本较高的违约客户,且能够实现分类变量的整组处理.与常规特征选择方法相比,w-LGL模型在数值模拟与实证研究中的分类效果更好.
Feature selection is an important step of credit score modeling.Reasonable feature selection method can not only simplify the structure of the model,but also improve the classification effect of the model.Based on the idea of weight adjustment of w-L1 SVM model and the advantage of selecting grouped variables of Logistic-Group-Lasso model,this paper proposes a weighted Logistic-Group-Lasso(w-LGL) model to select variables for credit score model.w-LGL model pays more attention to defaulting customers with high misclassification cost,and selects categorical variables by group.In numerical simulations and empirical analysis,w-LGL model performs better than other commom methods.
作者
夏利宇
姜天英
刘赛可
XIA Li-yu;JIANG Tian-ying;LIU Sai-ke(State Grid Energy Research Institute,Beijing 102209,China;Center for Applied Statistics,Renmin University of China,Beijing 100872,China)
出处
《数学的实践与认识》
北大核心
2020年第13期61-67,共7页
Mathematics in Practice and Theory
基金
教育部人文社会科学重点研究基地重大项目(15JJD910002)
国家电网有限公司管理咨询项目(SGTYHT/18-WT-236)。
关键词
信用评级模型
特征选择
权重调整
分组变量
credit score model
feature selection
weight adjustment
grouped variables
