基于高斯密度的一阶贝叶斯衍生分类器  

First-order Bayesian derivative classifier based on Gaussian density

作　　者：杜瑞杰[1] 王双成[1,2] 高瑞[2] 

机构地区：[1]上海立信会计学院数学与信息学院,上海201620 [2]上海立信会计学院立信会计研究院,上海201620

出　　处：《计算机应用研究》2015年第11期3242-3246,共5页Application Research of Computers

基　　金：国家自然科学基金资助项目(61272209);上海市自然科学基金资助项目(15ZR1429700);上海市教委科研创新项目(15ZZ099)

摘　　要：针对连续属性朴素贝叶斯分类器不能有效利用属性之间的条件依赖信息,而目前所进行的依赖扩展更关注效率,这使得扩展后分类器的分类准确性还有待提高等问题,使用高斯密度估计属性密度,将属性排序、分类准确性标准与属性父节点的贪婪选择结合,综合考虑效率和分类准确性,对朴素贝叶斯分类器进行依赖扩展,建立一阶贝叶斯衍生分类器,并对属性分类提供的信息进行分析。实验结果显示,基于高斯密度的一阶贝叶斯衍生分类器具有良好的分类准确性。Naive Bayesian classifier with continuous attributes can not effectively use conditional dependency intformatioo between attributes. At present, this paper emphasized the efficiency of its dependency extension, which made the classifi（ation accuracy of extended classifier need to be improved. Considering the efficiency and classification accuracy comprehensively and basing on Gaussian function to estimate attribute density, the dependency extension of naive Bayesian classifier with continuous attributes was clone by combining attribute sorting, classification accuracy criterion and the greedy selection of attribute parent node to establish first-order Bayesian derivative classifier. It also analyzed the information of attributes providing for class. Experiment results show that first-order Bayesian derivative classifiers have very good classification accuracy.

关 键 词：朴素贝叶斯分类器 高斯函数 贝叶斯网络 依赖扩展 分类准确性 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]