朴素贝叶斯判别的判别效用分析  

The Efficiency of Naive Bayes Classifier

作　　者：李永慈[1] 余欣宁[1] 王三强[1] 

机构地区：[1]北京林业大学理学院数学系,北京100083

出　　处：《生物数学学报》2010年第2期273-279,共7页Journal of Biomathematics

摘　　要：通过模拟实验研究了朴素贝叶斯判别的独立性假定对误判率的影响.实验结果表明:对于两个二维正态总体,当两类的距离为1、2、3时,在真实密度下进行判别的误判率随类内特征间相关系数的增加而减小,而朴素贝叶斯判别的误判率与特征间相关系数变化无关,因此随着类内特征间相关系数的增加,朴素贝叶斯判别的误判率明显增加,当类内特征间相关系数均为0.8时,朴素贝叶斯判别的误判率分别提高43.8%、220.8%和785.7%;对于高维正态总体模拟实验显示出同样的规律.The influence of features independence on the efficiency of naive bayes classifier is studied by simulation. Suppose there are two classes, each class has two features and Gaussian density. While the correlation between features is 0.8 and the distance between the two classes is 1, 2 and 3, the misclassification rate of naive bayes classifier has increased by 43.8%, 220.8% and 785.7%respectively.

关 键 词：核密度估计 贝叶斯判别 朴素贝叶斯判别 

分 类 号：S711[农业科学—林学]