关键词: 个人信用评估, 支持向量机, 改进DS证据理论, 属性约减

Abstract:

The effectiveness of credit scoring model will have a direct impact on credit institution’s gain and financial market’s order. In order to further improve the accuracy of credit evaluation model, an improving method of credit scoring combing Support Vector Machine（SVM） and improved DS evidence theory is proposed which also incorporates attribute reduction into modeling process. After reducing the redundancy attribute by C4.5 decision tree, the Basic Probability Assignment（BPA） of DS evidence is established based on the category label, confusion matrices and posterior probability of SVMs. Then it calculates the weight based on support between classifiers and expert to correct conflict evidence which produced during the process of training SVMs. Finally it makes final decision according to improved DS evidence theory and SVM ensemble. It is proved that the proposed method is stable, highly accurate, strong robust and feasible. It can be a useful tool for credit scoring.

Key words: personal credit scoring, support vector machine, improved DS evidence theory, feature reduction