基于多分类器串并联结构的个人信用评估模型  

Personal credit evaluation model based on multi-classifier series and parallel structure

作　　者：杨柳 孙带[2] YANG Liu;SUN Dai(National Center for Applied Mathematics in Hunan,Xiangtan 411105,China;School of Mathematics and Computational Science,Xiangtan University,Xiangtan 411105,China)

机构地区：[1]湖南国家应用数学中心,湖南湘潭411105 [2]湘潭大学数学与计算科学学院,湖南湘潭411105

出　　处：《湘潭大学学报（自然科学版）》2022年第6期1-11,共11页Journal of Xiangtan University(Natural Science Edition)

基　　金：国家自然科学基金面上项目(12071399);湖南省学位与研究生教育改革研究项目(2019JGYB109);湖南国家应用数学中心项目(2020ZYT003)。

摘　　要：为了降低因贷款者违约给金融机构带来的损失和风险,对贷款者的个人信用进行评估有着极其重要的意义.针对贷款数据集的不平衡和高维特征问题,结合集成学习中串联与并联结构的优点,提出基于多分类器串并联结构的个人信用评估模型.利用SMOTE算法平衡数据集正负样本个数、随机森林计算特征重要性并筛选特征,再在新数据集上训练K近邻、Logistic回归、决策树、支持向量机4个基分类器,将它们的预测概率采用简单平均法进行综合,这样就形成了随机森林与4个基分类器的串并联结构.通过在真实数据集上训练和测试,发现该模型在预测准确率和区分好坏样本上都有显著优势,具有一定的适用性.In order to reduce the losses and risks caused by the default of the lenders to the financial institutions,it is of great significance to evaluate the personal credit of the lenders.Aiming at the problem of unbalanced and high-dimensional characteristics of loan data sets,combining the advantages of series and parallel structure in ensemble learning,a personal credit evaluation model based on multi-classifier series and parallel structure is proposed.SMOTE algorithm is used to balance positive and negative sample number of data sets,and random forest is used to calculate the importance of features and screen features,then train four base classifiers,K-nearest neighbor,Logistic regression,decision tree and support vector machine on the new data sets,and integrate their prediction probability using simple average method,thus a series and parallel structure of random forest and four base classifiers is formed.Through training and testing of real data sets,it is found that the model has significant advantages in predicting accuracy and distinguishing good and bad samples,and has certain applicability.

关 键 词：个人信用评估 SMOTE算法 随机森林 串并联结构 

分 类 号：O152.1[理学—数学]