一类泊松-多项模型参数估计的EM算法.PDFVIP

第11卷第19 期 Vol.11 No.19 2018 年 10 月 October 2018 一类泊松-多项模型参数估计的 EM 算法 李 苗，魏立力* （宁夏大学数学统计学院，银川 750021 ） 摘要：研究一类泊松-多项模型参数估计的期望最大化（expectation maximization ，EM ）算法。首先，针对完 全数据情形推导出了极大似然估计（maximum likelihood estimate ，MLE ）的解析表达式；其次，设计适用于 缺失数据情形参数估计的 EM 算法，给出了算法迭代序列的表达式；最后，通过一个算例说明了所提方法的 可行性。结果表明，EM 算法解决了极大似然估计法无法得到缺失数据对应参数估计值的问题，适用于不完全 数据情形的泊松-多项模型参数估计，体现了一定的优越性。 关键词：数理统计学；EM 算法；缺失数据；泊松-多项分布 中图分类号：O212 文献标识码：A 文章编号：1674-2850(2018)19-1943-05 EM algorithm for parameter estimation of a class of Poisson-multinomial models LI Miao, WEI Lili (School of Mathematics and Statistics, Ningxia University, Yinchuan 750021, China) Abstract: An expectation maximization (EM) algorithm for parameter estimation of a class of Poisson-multinomial models is studied in this paper. Firstly, the analytical expression of maximum likelihood estimate (MLE) is derived for the complete data case. Then the EM algorithm for parameter estimation is designed for the missing data case, and the expression of iterative sequence of the algorithm is given. Finally, an example is given to illustrate the feasibility of the proposed method. The results show that the EM algorithm not only solves the problem that the maximum likelihood estimation method cannot obtain the estimated value of corresponding parameters in the missing data, but also applies to the parameter estimation of Poisson-multinomial models for the incomplete data case, so that the superiority of the EM algorithm is reflected. Key words: mat