Bayes判别判别效果的模拟研究  

Stimulated Research on the Effect of Bayes Discriminant

作　　者：曾澄波[1] 缪秋萍 郜艳晖[1] 李丽霞[1] 

机构地区：[1]广东药学院公共卫生学院,广州510310

出　　处：《数理医药学杂志》2016年第9期1272-1275,共4页Journal of Mathematical Medicine

基　　金：广东药学院创新强校工程资助项目;2013年广东省学位与研究生教育改革研究项目(2013JGXM-MS42);2014年广东省高等教育教学改革项目(GDJG20141169);广东省创新强校工程应用型人才培养示范专业"公共卫生与预防医学"基金资助项目(粤教高函[2014]97号);"广东省预防医学专业综合改革试点项目";"预防医学人才协同创新培养模式探索与实践"基金资助项目(粤教高函[2013]113号)

摘　　要：目的:通过计算机模拟实验探讨Bayes判别的判别效果。方法:模拟类内服从多元正态分布的数据在不同类别数、样本量、变量个数、均值向量、方差协方差矩阵、先验概率的情况下Bayes判别的误判率,评价判别效果。结果:不同类别数、变量个数、均值向量、有无先验概率均能影响误判率;不同的样本量和方差虽然能够改变误判率,但误判率间差异无统计学意义。结论:计算机模拟实验是演示统计模型中关键概念以及直观的体验统计模型中参数间相互间关系的理想工具。Objective To study the effect of Bayes discriminant in simulated data. Methods： The simulated data was generated from multivariate normal distribution. According to the different classes, sample sizes , numbers of variables,mean vectors,variance covariance matrixes and priors,the effect of Bayes discriminant was assessed. Results： Different classes,numbers of variables and mean vectors all can change the false rates of Bayes discriminant. So do the priors. However, although different sample sizes and variances can also change the false rate of Bayes discriminant, the difference of false rates has no statistical significance. Conclusion： Computer simulation experiment is an ideal tool to demonstrate the key concepts and the relationship among the parameters in the statistical model.

关 键 词：Bayes判别 模拟研究 误判率 

分 类 号：R195.1[医药卫生—卫生统计学]