摘要
为了改善树增强朴素贝叶斯(TAN)的分类精度,对TAN结构进行了扩展,提出了一种利用可分解的评分函数构建树形贝叶斯网络分类模型的学习方法。在构建TAN网络时允许属性没有父结点。采用低阶CI测试初步剔除无效属性,再结合改进的BIC评分函数利用贪婪搜索获得每个属性结点的父结点,从而建立分类模型。对比朴素贝叶斯(NB)和TAN,提出的分类算法在分类准确率和AUC面积两个指标上表现更好,说明本文模型拥有比TAN更好的分类效果。
In order to improve the classification accuracy of Tree Augmented Naive Bayes(TAN),extended the TAN structure, and proposed a learning method of building the tree-structure Bayesian network classifier with using a decomposable scoring function. When constructing a classification network,allows each attribute to have no parent node or only one attribute parent node. Applies the low-order CI test to eliminate the useless attribute firstly,and then based on improved BIC function,acquires the parent node of each attribute node with the greedy algotithom,to establish the classification model. The modified classifier behaves better than NB and TAN on both the classification accuracy and the AUC area,which proves that the proposed model performs better than TAN.
作者
张坤
陈曦
宋云
傅明
ZHANG Kun;CHEN Xi;SONG Yun;FU Ming(College of Computer and Communication Engineering,Changsha University of Science and Technology,Changsha,Hunan 410114,China)
出处
《计算技术与自动化》
2019年第1期55-61,共7页
Computing Technology and Automation
基金
国家自然科学基金资助项目(61772087)
长沙理工大学研究生科研创新项目(CX2017SS20)
关键词
树增强朴素贝叶斯
分类网络
评分函数
tree augment Naive Bayes
classification network
scoring function