新型加权粗糙朴素贝叶斯算法及其应用研究  

Novel weighted rough naive Bayes algorithm and its application

作　　者：王辉[1] 黄自威 刘淑芬[2] 

机构地区：[1]河南理工大学计算机科学与技术学院,河南焦作454000 [2]吉林大学计算机科学与技术学院,长春130012

出　　处：《计算机应用研究》2015年第12期3668-3672,3692,共6页Application Research of Computers

基　　金：国家自然科学基金资助项目(51174263;61300216);国家教育部博士点基金资助项目(20124116120004);河南省教育厅科学技术研究重点项目(13A510325)

摘　　要：待分类数据集中通常存在大量的冗余属性,会严重影响分类效率。为了达到在降低计算复杂度的同时提高分类准确率的目的。首先在朴素贝叶斯模型中引入粗糙集技术对数据集进行属性约简,获取最优属性子集;然后在此基础上以最大化数据集的对数条件似然估计为标准对条件属性设定(近似)最优权值,进而提出一种新型加权粗糙朴素贝叶斯模型。通过在垃圾邮件过滤领域对该模型进行实际验证表明,贝叶斯模型的分类效率有明显提高,而且分类性能更加稳定,证明该方法不仅可以有效去除冗余属性,而且为条件属性赋予的权值较之传统加权方法更加合理。There are usually a lot of redundant attribute stay in data set, which can seriously influence efficiency of dataset classfication. In order to achieve the goal of reducing the computational complexity, and improving the accuracy of classification at the same time, this paper introduced rough set technology into naive Bayes model for attribute reduction to obtain the optimal attributes subset. Then, on this basis, took conditional of logarithmic likelihood estimation of data set as standard to set the （approximate） optimal weights for the attribute, and proposed a novel kind of weighted rough naive Bayes model. The practical classification performance of this model in the filed of spare filter show that the classification efficiency of the Bayes model is obviously improved, and the classification performance is more stable, which proves that the method can eliminate the redundant attributes effectively, and the weight value that set for attribute is more reasonable than the traditional weighted method.

关 键 词：加权朴素贝叶斯 决策表 属性约简 对数条件似然估计 垃圾邮件过滤 

分 类 号：TP301.6[自动化与计算机技术—计算机系统结构]