基于贝叶斯网络的SVM客户信用评估模型研究

发布时间：2018-06-09 14:32

本文选题：贝叶斯网络 + 支持向量机　；参考：《辽宁工程技术大学》2017年硕士论文

【摘要】：受数据收集条件所限,加上数据录入时出现数据遗漏等情况,银行客户信用评估中经常会出现数据严重缺失。这些数据的缺失会对银行客户信用评估造成极大影响。对缺失值的处理常用方法有删除含缺失数据的记录或者对缺失数据进行填充等方法,显然,缺失数据填充将为未来信用评估奠定基础。本文建立基于贝叶斯网络与概率推理相结合的缺失数据填充方法。根据已有数据信息,挖掘属性之间的相关性,建立贝叶斯网络,由于贝叶斯网络的特性可以简化多维联合概率的求解。根据概率推理的结果对缺失数据进行填充,推理结果还能反映出数据填充的精度,即概率越高,数据填充的准确率越高。在利用贝叶斯网络推理方法处理缺失数据的基础上,建立基于支持向量机的客户信用评估体系,该体系包含三阶段的信用评估模型,即是否违约的判定模型、违约概率的计算模型以及客户信用风险度的预测模型。信用评估体系的建立,拓宽了客户信用评估的辐射面,可以更加全面的对客户信用进行评估,为银行提供决策依据,降低银行的不良信贷率,具有重要的理论以及现实意义。
[Abstract]:Limited by the condition of data collection and the omission of data in data entry, the bank customer credit evaluation often has serious data deficiency. The lack of these data will have a great impact on the credit evaluation of bank customers. The common methods to deal with missing values are to delete records containing missing data or fill in missing data. Obviously, the filling of missing data will lay a foundation for future credit evaluation. In this paper, the missing data filling method based on Bayesian network and probabilistic reasoning is established. Based on the existing data information, the correlation between attributes is mined, and Bayesian networks are established. Because of the characteristics of Bayesian networks, the solution of multidimensional joint probability can be simplified. According to the result of probabilistic reasoning, the missing data is filled, and the result can reflect the precision of data filling, that is, the higher the probability, the higher the accuracy of data filling. On the basis of using Bayesian network reasoning method to deal with missing data, a customer credit evaluation system based on support vector machine (SVM) is established. The system consists of a three-stage credit evaluation model, that is, the judgment model of default or non-default. The calculation model of default probability and the prediction model of customer credit risk. The establishment of credit evaluation system broadens the radiation surface of customer credit evaluation, which can evaluate customer credit more comprehensively, provide decision basis for banks, and reduce the bad credit rate of banks, which has important theoretical and practical significance.
【学位授予单位】：辽宁工程技术大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：F274;TP18

【参考文献】