基于粗糙集属性约简与进化算法的贝叶斯网络分类器  

Bayesian Network Classifiers Based on Attribute Reduction and Evolutionary Algorithm

作　　者：李艳颖[1,2] 杨有龙[2] 汪春峰[3] 

机构地区：[1]宝鸡文理学院数学系,陕西宝鸡721006 [2]西安电子科技大学数学与统计学院,陕西西安710071 [3]河南师范大学数学与信息科学学院,河南新乡453007

出　　处：《郑州大学学报（理学版）》2014年第2期43-49,共7页Journal of Zhengzhou University:Natural Science Edition

基　　金：国家自然科学基金资助项目;编号61075055;宝鸡文理学院重点项目;编号ZK10107;西安电子科技大学基本科研业务资助项目;编号K5051270013;宝鸡市项目;编号2013RT-3;陕西省青年基金项目;编号2014JQ1020

摘　　要：提出一种新的学习无约束贝叶斯网络分类器的算法(RE-BNC).该算法基于粗糙集理论,在保证分类精度不变的前提下,先对冗余属性变量进行约简,降低属性变量维数,然后构建一个无约束优化模型用来学习较好的初始种群,降低搜索空间,再结合进化算法学习分类器的网络结构.与其他常见的8种分类器算法相比较,实验结果表明该算法设计合理,且分类效果较好.A new algorithm for learning Bayesian network classifiers based on rough set approach and evolutionary algorithm（RE-BNC） was proposed.Firstly,to reduce the dimensions of the attribute variables,redundant attributes of the to-be-classified data set were removed according to the rough set theory,and the accuracy was nondecreasing.Then an unconstrained optimization model was constructed for learning a better initial population and reducing the search space.At last,the structure of the classifier was learned by taking advantage of the evolutionary algorithm.Compared with other 8 kinds of classifier algorithms,the results showed that RE-BNC was reasonable and had higher accuracy.

关 键 词：贝叶斯网络 分类器 进化算法 属性约简 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]