C5.0分类算法及在银行个人信用评级中的应用

研究了商业银行的个人信用评级问题.由于个人信用记录中既涉及有数值型数据, 也涉及有非数据型数据,而决策树是解决这一类问题的最好方法. 到目前为止,决策树C4.5算法的研究已基本成熟,但其C5.0算法由于商业机密的缘故至今没有公开.因此在决策树C4.5算法基础上详细研究了C5.0算法及相应的Boosting技术,并嵌入Boosting算法技术, 构造了成本矩阵和Cost-sensitive tree,以此建立基于C5.0算法的银行个人信用评级模型,用来对德国某银行的个人信贷数据进行信用评级,同时对模型参数调整前后决策树的判别结果进行比较. 仿真表明:参数调整后的决策树的判别结果优于参数调整前的决策树的判别结果.

C5.0 classification algorithm and its application on individual credit score for banks

This paper mainly focuses on individual credit evaluation of a commercial bank. The records of individual credit include numerical value as well as un-numerical values. Decision tree is a good solution for this kind of issue. Until now, research on the decision tree algorithm of C4.5 has become mature, but C5.0 algorithm is still not open because of commercial secret. This article does some detailed research into C5.0 algorithm and its related technology of "boosting". We also construct cost matrix and cost-sensitive tree, embed with boosting technology, and finally establish the individual credit evaluation model of Commercial Bank based on C5.0, which is used to evaluate the individual credit records of a German bank. Meanwhile, we compare the adjusted decision tree model and the original one. The simulation result shows that the evaluation result of the adjusted decision tree model is better.