扩展的树增强朴素贝叶斯网络信用评估模型  

The credit scoring model on extended tree augment naive Bayesian network

作　　者：李旭升[1] 郭春香[2] 郭耀煌[1] 

机构地区：[1]西南交通大学经济管理学院,成都610031 [2]西南科技大学理学院,绵阳621000

出　　处：《系统工程理论与实践》2008年第6期129-136,共8页Systems Engineering-Theory & Practice

基　　金：国家自然科学基金(70371026,70771093);四川省教育厅科研项目(2006C082)

摘　　要：针对信用评估问题的特点,在推导混合数据极大似然函数的基础上,提出了扩展的树增强朴素贝叶斯网络信用评估模型,用10层交叉验证在真实数据集上进行了测试并与神经网络分类模型进行了比较.测试结果表明扩展的树增强朴素贝叶斯网络信用评估模型具有较高的分类精度,在信用评估中具有优势.Aiming to the specialty of credit scoring, the maximum likelihood function of hybrid data is deduced, and a new credit scoring model called Extended Tree Augmented Naive Bayesian Classifier is put forward. The proposed model is tested using 10-fold cross validation with a real world data set, and compared with neural network models. Results demonstrate that the credit scoring model on extended tree augmented naive Bayesian classifier is competitive with neural network models and predominant in credit scoring domain

关 键 词：信用评估 贝叶斯网络 树增强朴素贝叶斯分类模型 扩展的树增强朴素贝叶斯分类模型 神经网络分类模型 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]