基于Tri-training算法的多分类信用评级方法

Multi-class credit rating method based on the Tri-training algorithm

摘要:

随着经济的快速发展，信用贷款在企业资金周转中的作用越来越重要. 信用评级是信用贷款发放的基本依据之一. 本文针对实际信用评级中有标签样本数量不足的问题提出了一种基于Tri-training算法的多分类信用评级方法, 该方法选择支持向量机、决策树和最大熵模型作为基分类器组合. 最后，本文使用真实信用数据集验证了该方法的实际效果.

Abstract:

With the rapid development of economy credit loans become more and more imporant in the capital turnover of corporations. Credit rating is a base of credit loan. In this paper, we focus on the problem of insufficient number of label samples in actual credit rating and propose a multi-class credit rating method based on the Tri-training algorithm, which selects the support vector machine, the decision tree and the maximum entropy model as the base classifiers combination. Finally, the performance of the method is verified by using some real credit datasets.