摘要
首先提出了处理0和1数据偏多的零一膨胀泊松回归模型,其次对模型建立了参数的极大似然估计.针对传统的EM算法只能使得估计收敛到局部极大值这个缺陷,提出了一种随机EM算法对传统的EM算法进行修正,使得模型能够找到全局最优解.最后通过模拟研究说明该方法的有效性.
This paper proposes zero-and-one inflated Poisson regression model and gives the maximum likelihood estimation of parameters.Since standard EM algorithm will make the parameter estimate converge to local maxima,to address this shortcoming,this paper introduces the traditional EM algorithm amendment proposal which is called stochastic EM algorithm.The algorithm makes it possible to find the global optimum.Finally,the corresponding simulation is given to illustrate the proposed method.
出处
《河南科学》
2017年第7期1037-1041,共5页
Henan Science
基金
国家自然科学基金项目(11647050)
关键词
零一膨胀
计数数据
随机EM算法
zero-and-one inflation
count data
stochastic EM algorithm