一种基于概率加权的朴素贝叶斯分类  

A Naive Bayesian Classification Model Based on the Probabilistic Weights

作　　者：白似雪[1] 梅君[1] 吴穹[2] 朱涛[1] 

机构地区：[1]南昌大学计算机科学与技术系,江西南昌330031 [2]北京邮电大学国际学院,北京102209

出　　处：《南昌大学学报（理科版）》2009年第2期191-194,共4页Journal of Nanchang University(Natural Science)

基　　金：国家自然科学基金资助项目(60863002)

摘　　要：朴素贝叶斯分类是一种简单而高效的方法,但是它的属性独立性假设,影响了它的分类性能。为了克服该问题,提出了一种基于概率推理的加权朴素贝叶斯分类模型。通过计算属性和类之间的相关概率和不相关概率,对属性赋予不同的权重,从而在保持简单性的基础上有效地提高了朴素贝叶斯算法的分类性能。实验结果表明,该方法可行而且有效。Naive Bayes Classifier is a simple and effective classification method, but its attribute independence as- sumption makes it unable to express the dependence among attributes in the real world, and affects its classification performance. In this paper a method for setting attribute weights based on probabilistic inference for using with Naive Bayes is presented. It improve the classification performance of Naive Bayes through set attributes different weights by their related - probability and unrelated - probability with classes. Experimental results illustrate the effi- ciency of this method.

关 键 词：数据挖掘 朴素贝叶斯 概率推理 相关概率 

分 类 号：TP274.3[自动化与计算机技术—检测技术与自动化装置]