后验概率支持向量机在企业信用评级中的应用

Abstract

摘要:在支持向量机 ( Support Vect orMachine)的分类问题中,训练样本的分类信息总是确定的,由此得到的分类指示函数也总是对新样本给出确定的分类信息,但是这种情况对一些不确定性问题并不恰当。利用贝叶斯规则,将样本的后验概率与传统支持向量机结合,得到了基于后验概率的支持向量机。在具体的算法上,引入了一个经验性的方法得到样本的后验概 率。以某评级机构提供的企业信用评估数据库为研究对象。 ABSTRACT: The classified information of the training sample is always certain in the classification problem of support vector machine . The indicat or function obtained always gives a certain classificati on information to the new sample . But it is not appropriate to some uncertain problems . This paper obtains the SVM based on posteriori probability by utilizing the Bayes rule to combine posteriori probability with SVM. An experiential manner is proposed to estimate the posteriori probability of the training data .