基于Tri-training算法的多分类信用评级方法  

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

作　　者：曹欣妍 周杰[1] CAO Xin-Yan;ZHOU Jie(School of Mathematics,Sichuan University,Chengdu 610064,China)

机构地区：[1]四川大学数学学院,成都610064

出　　处：《四川大学学报（自然科学版）》2023年第2期13-18,共6页Journal of Sichuan University(Natural Science Edition)

基　　金：国家自然科学基金(11871357)。

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,本文使用真实的信用数据集验证了该方法的实际效果.Credit loans become more and more important in the capital turnover of corporations with the rapid development of economy.Credit rating is the base of credit loan.In this paper we focus on the problem of insu icient 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.

关 键 词：多分类信用评级 半监督学习 TRI-TRAINING 

分 类 号：O175.5[理学—数学]