基于非参数K近邻-秩相关的信用评价指标构建模型  

Construction Model of Credit Evaluation Index Based on Nonparametric K Neighborhood-Rank Correlation

作　　者：王璐[1] 刘妍[2] 李战江[3] 郭林 WANG Lu;LIU Yan;LI Zhanjiang;GUO Lin(College of Economics and Management,Beijing Forestry University,Beijing 100083,China;Chifeng Qiaobei Sub-branch,Industrial and Commercial Bank of China,Chifeng 024050,China;College of Economics and Management,Inner Mongolia Agricultural University,Hohhot o10010.China)

机构地区：[1]北京林业大学经济管理学院,北京100083 [2]中国工商银行股份有限公司赤峰桥北支行,内蒙古赤峰024050 [3]内蒙古农业大学经济管理学院,呼和浩特010010

出　　处：《内蒙古大学学报（自然科学版）》2021年第6期570-577,共8页Journal of Inner Mongolia University：Natural Science Edition

基　　金：国家自然科学基金项目(72161033);内蒙古自然科学基金项目(2020MS07009)。

摘　　要：针对现有研究中指标的遴选标准与违约状态的鉴别能力无关且遴选出的指标信息冗余的不足,提出了一种将非参数K近邻判别与spearman秩相关分析相结合的信用评价指标体系构建模型,并且使用经营家庭农牧场的农户信贷数据进行了实证分析。创新与特色:一是通过非参数K近邻模型确定特定指标对于农户的判别精度,根据每一个指标判别精度的大小,筛选出对农户具有显著鉴别能力的指标;二是通过spearman秩相关分析计算经过第一轮指标筛选保留下来的指标的spearman秩相关系数,在秩相关系数大于等于0.7的一对指标中删除判别精度较小的那个指标,避免了评价指标体系的信息冗余。检验结果表明:一是ROC曲线的有效性检验揭示出最终构建的农户信用评价指标体系是有效的;二是利用非参数K-近邻模型筛选出的指标体系在样本判别精度方面要显著优于传统贝叶斯模型。In view of the deficiency that the selection criteria of indicators in the existing research have nothing to do with the ability to identify default status and the selected indicators have information redundancy, a credit evaluation indicator system construction model of farmers from family farm and family ranch is proposed combining the non-parameter K-neighbor discrimination and Spearman rank correlation analysis.The innovation and characteristics of this paper are as follows: the first point is to use the non-parametric K-nearest neighbor model to calculate the discriminant accuracy of the specific index.According to the size of the discriminant accuracy of each index, the indicators that truly discriminate the samples are selected.The second point is to use the spearman rank correlation analysis to calculate the spearman rank correlation coefficient of the indicators that have been retained through the first round of index selection, and to delete the index with a smaller judgment accuracy in a pair of indexes with a correlation coefficient>0.7.The test results show that: Firstly, the validity test of the ROC curve reveals that the constructed credit index system for rural households is effective.Secondly, the index system selected by the non-parametric K-nearest neighbor model is significantly superior to the traditional bayesian model in terms of sample discrimination accuracy.

关 键 词：信用评价 非参数K近邻判别 spearman秩相关分析 

分 类 号：O212.7[理学—概率论与数理统计] F832.43[理学—数学]