基于BP神经网络和Logistic回归的农户信用评价研究

发布时间：2018-04-02 10:08

本文选题：农户信用评分　切入点：农户信用风险　出处：《湖南大学》2012年硕士论文

【摘要】：本文的研究目的就是通过对申贷农户的各项指标的分析，找到能够显著区分信用等级较高的农户(好客户)和信用等级较低的农户(坏客户)的指标，通过构建模型计算农户为好客户的概率，进而得到信用得分，最终得到农户信用等级评判标准，为有关各方提供决策依据和参考意见。本文首先给出了目前湖南省农信社普遍采用的农户信用评级方法，然后对当前评定过程中存在的问题进行了分析，在借鉴国内外研究成果的基础上，通过比较信用评价的各种方法，同时结合我国农户的信用特征，最终选择基于BP神经网络和Logistic回归两种方法，并且综合考虑两种评级方法的优缺点，构建基于两种方法的混合模型，提高预测精度和稳定性。接下来，在分析农户信用风险的产生及其风险独特性的基础上，得出了我国农户信用特征是个人和中小企业的结合体。因此，在指标体系的构建上可以参考国内外的关于个人和中小企业的信用评价指标体系。以我国湖南部分地区申请贷款的农户为研究对象，分别从农户户主及家庭成员情况、资产情况、负债情况、经营情况、家庭开支五个方面构建初始评价指标，总共选取22个指标，借助SPSS统计分析软件，利用因子分析法进行分析提取了12个主要指标，总共选取了646户农户家庭作为样本，在此基础上分别比较基于BP神经网络的农户信用评价模型和基于Logistic回归的农户信用评价模型的应用效果，然后构建基于二者的组合模型，实证结果显示：该模型的总体准确率为97.1%，其中将好客户判断为好客户的准确率为98.8%，将坏客户判断为坏客户的准确率为81.8%，可解释性及稳健性都是比较理想的。可见，，此模型取得了较好的预测效果，具有一定的应用价值。
[Abstract]:The purpose of this paper is to find out the index that can distinguish the high credit grade farmer (good customer) and the low credit grade farmer (bad customer) through the analysis of the indexes of the farmers applying for loans. The probability of farmers being a good customer is calculated by constructing a model, and then the credit score is obtained. Finally, the evaluation standard of the farmer's credit grade is obtained, and the decision basis and reference advice are provided for the parties concerned. At first, this paper gives the credit rating method of farmers, which is widely used in Hunan Rural Credit Cooperative, then analyzes the problems in the current evaluation process, and draws lessons from the domestic and foreign research results. By comparing various methods of credit evaluation and combining the credit characteristics of farmers in China, two methods based on BP neural network and Logistic regression are selected, and the advantages and disadvantages of the two rating methods are considered synthetically. A hybrid model based on two methods is constructed to improve prediction accuracy and stability. Then, on the basis of analyzing the emergence and uniqueness of peasant household credit risk, it is concluded that the characteristics of peasant household credit in China are the combination of individuals and small and medium-sized enterprises. In the construction of the index system, we can refer to the credit evaluation index system of individuals and small and medium-sized enterprises at home and abroad. This paper constructs the initial evaluation index in five aspects of assets, liabilities, operation and household expenditure. In total, 22 indexes are selected, and 12 main indexes are extracted by factor analysis with the help of SPSS statistical analysis software. A total of 646 households were selected as samples, and then the application effects of the model based on BP neural network and the model based on Logistic regression were compared, and then the combination model based on the two models was constructed. The results show that the overall accuracy of this model is 97.1g, in which the accuracy of judging good customers as good customers is 98.8, and that of bad customers is 81.8. This model has achieved good prediction effect and has certain application value.
【学位授予单位】：湖南大学
【学位级别】：硕士
【学位授予年份】：2012
【分类号】：F832.43;F224

【参考文献】