决策树算法在农村信用社农户信用评级中的应用

发布时间：2018-01-18 12:04

本文关键词：决策树算法在农村信用社农户信用评级中的应用　出处：《湖南大学》2013年硕士论文　论文类型：学位论文

【摘要】：农村信用合作社（农村合作银行、农村商业银行，以下简称“农信社”）长期以来作为农村金融的主力军，在服务农业生产者和小商品生产者，支持农业发展方面发挥了无可替代的作用，在解决农户贷款难，促进农民增收,支持农村经济发展和农业现代化等方面发挥了重要而积极的作用，但其自身不良贷款率仍然是商业银行的5-7倍。因此，客观、全面、准确的评估农村客户的还款能力和还款意愿，拒绝不符合条件的客户，将是避免、控制、减少损失的一个重要手段。传统的信用评分由于标准不一，具有成本高、主观性强、效率低下的特点。社会经济的发展，农户概念的外延和内涵都发生了较大改变，再用传统的农户信用评价体系来运用到现在的农户概念上，不合时宜的。欧美国家的使用经验表明，个人信用评分具有快速处理客户贷款申请，而且成本低、标准一致，比较客观的特点，在银行信用风险管理中发挥重要的作用。欧美国家现代信用评分体系广泛运用了统计学、运筹学、人工智能等方面的技术，在此基础上形成的数据挖掘技术在信用评分模型的构建中发挥着广泛而且重要的作用。本文首先介绍了国内外学者研究个人信用评级的不同理论与方法，并简要介绍了其优缺点；其次分析了农户信用评级现状，定性评价在评级过程中所占比重较大，因而建立一种定量的评级方法对降低贷款风险具有重要意义；第三，介绍了本文研究所采用的的决策树技术、数据挖掘方法以及评级工具——SAS；第四，本文收集了本银行系统近4年的真实农户贷款样本数据，利用SAS中的决策树算法，通过数据清洗、转换，抽样、分析，建立决策树模型，并对属性进行赋值，，建立农户信用评分模型；第五，本文得出的农户信用评分模型采用百分制，不同的分值对应相应的信用等级，从而采取不同的信贷策略。本文创新点在于将传统信用评级的定量指标由占比不到70%提高到94%，大大提升了农户信用评级的精确度，同时测试得出农户信用评级模型在高信用级别的客户中具有较高的预测精度，对于中、低级信用级别的客户精度还有待改善的观点。
[Abstract]:Rural credit cooperatives (rural cooperative banks, rural commercial banks, hereinafter referred to as "rural credit cooperatives") have long been the main force in rural finance, serving agricultural producers and small commodity producers. It has played an irreplaceable role in supporting agricultural development and has played an important and active role in solving the difficulties of farmers' loans, promoting farmers' income, and supporting rural economic development and agricultural modernization. But its own non-performing loan rate is still 5-7 times of commercial banks. Therefore, objective, comprehensive and accurate evaluation of rural customers' repayment ability and repayment will be avoided and controlled. The traditional credit rating has the characteristics of high cost, high subjectivity, low efficiency and the development of social economy. The extension and connotation of the concept of peasant household have changed greatly. It is inappropriate to use the traditional credit evaluation system of farmers to the present concept of farmers. The experience of European and American countries shows that. Personal credit rating has the characteristics of rapid processing of customer loan applications, low cost, consistent standards, and more objective characteristics. In the bank credit risk management plays an important role. Europe and the United States modern credit scoring system has been widely used in statistics, operational research, artificial intelligence and other aspects of technology. On this basis, data mining technology plays an important and extensive role in the construction of credit scoring model. This paper first introduces the different theories and methods of personal credit rating, and briefly introduces its advantages and disadvantages. Secondly, the paper analyzes the present situation of farmers' credit rating, and the qualitative evaluation occupies a large proportion in the process of rating, so it is of great significance to establish a quantitative rating method to reduce the risk of loans. Thirdly, this paper introduces the decision tree technology, data mining method and rating tool, which are used in this paper. In 4th, this paper collects the real farmer loan sample data of the bank system in the past 4 years, using the decision tree algorithm in SAS, through data cleaning, transformation, sampling, analysis, building decision tree model. The attribute is assigned and the credit scoring model is established. In 5th, the credit rating model of farmers adopted the percent system, and the different scores correspond to the corresponding credit grade, thus adopting different credit strategies. The innovation of this paper lies in increasing the quantitative index of traditional credit rating from less than 70% to 94, which greatly improves the accuracy of farmers' credit rating. At the same time, it is concluded that the farmer credit rating model has high prediction accuracy among high credit grade customers, and the customer accuracy of middle and low credit grade still needs to be improved.
【学位授予单位】：湖南大学
【学位级别】：硕士
【学位授予年份】：2013
【分类号】：F832.43;TP311.13

【参考文献】