带有测量误差Poisson回归模型的变量选择  

Variable Selection in Poisson Regression Model with Measurement Error

作　　者：刘芳 LIU Fang(Registrar's Office,Gannan Normal University,Ganzhou 341000,China)

机构地区：[1]赣南师范大学教务处,江西赣州341000

出　　处：《赣南师范大学学报》2024年第6期44-47,共4页Journal of Gannan Normal University

基　　金：江西省教育厅科技项目(GJJ211403)。

摘　　要：Poisson回归模型是一种特殊的广义线性模型,被广泛用于计数数据的建模.然而在很多应用中,模型中的协变量很难被精确测量,在进行变量选择和参数估计时,这些测量误差通常难以处理,特别是当维数较高时.针对这一问题,本文提出了一种惩罚的偏差校正方法以确定带有测量误差的Poisson回归模型(MPR)的真实结构.该方法惩罚了专门设计用于处理测量误差的目标函数,并在一个有界的区域内利用复合梯度下降算法给出回归参数的估计.通过随机模拟验证了本文所提出方法的有限样本性质.Poisson regression is a special generalized linear model which is widely used to model count data.However,in many applications,covariates are often contaminated with errors.Errors in these covariates are usually difficult to handle,especially when the covariate dimension is high.This paper proposes a bias-corrected penalized method to determine the underlying structure of Poisson regression model with measurement errors(MPR).The procedure penalizes a target function that is specifically designed to handle measurement errors and provides the composite gradient descent algorithm within a bounded region.The numerical performance is demonstrated using simulation studies.

关 键 词：Poisson回归模型 测量误差 变量选择 惩罚函数 

分 类 号：O212.1[理学—概率论与数理统计]