基于Dirichlet分布有限混合模型的Bayes聚类  

Bayesian Clustering Based on Finite Mixture Models of Dirichlet Distribution

作　　者：俞燕[1] 徐勤丰[1] 孙鹏飞[1] 

机构地区：[1]复旦大学统计学系,上海200433

出　　处：《应用数学》2006年第3期600-605,共6页Mathematica Applicata

基　　金：国家自然科学基金资助项目(10271078)

摘　　要：本文基于Dirichlet分布有限混合模型,提出了一种用于成分数据的Bayes聚类方法.采用EM算法获得模型参数的估计,用BIC准则确定类数,用类似于Bayes判别的方法对各观测分类.推导了计算公式,编写出程序.模拟研究结果表明,本文提出的方法有较好的聚类效果.Based on finite mixture models,we propose a Bayesian clustering method for compositional data. EM algorithm is adopted to compute the estimates of model parameters, BIC is employed to determine the number of clusters,and an analogue procedure of Bayesian discrimination is used to classify each observation. We also deduce iteration formula and write related program. The simulation study shows that this method works well with acceptable clustering results.

关 键 词：成分数据 Bayes聚类 Dirichlet分布 有限混合模型 EM算法 BIC准则 

分 类 号：O212.4[理学—概率论与数理统计]