统计与决策2024，Vol.40Issue(12)：46-51,6.DOI:10.13546/j.cnki.tjyjc.2024.12.008
超高维纵向数据部分线性模型的特征筛选
Feature Screening for Partially Linear Models With Ultrahigh-dimensional Longitudinal Data
摘要
Abstract
Feature screening of ultrahigh-dimensional longitudinal data is one of the difficulties of ultrahigh-dimensional feature screening,and the difficulty is to estimate the working correlation coefficient matrix under the premise of ensuring the speed of marginal screening.Under the assumption of partial linear model,this paper takes into account the characteristics of lon-gitudinal data of between-group independence and within-group correlation,and uses sample covariance to estimate the un-known working covariance matrix,proposing a sure independent screening method with profile covariance matrix(PMSIS).The paper also theoretically proves that the method has the sure screening property under certain regularity conditions,and verifies the finite sample properties of the method through Monte Carlo numerical simulation and gut microbiota data.The results show that the new PMSIS method can be used to effectively screen weakly correlated covariates.关键词
超高维/纵向数据/部分线性模型/特征筛选/确定筛选性质Key words
ultrahigh dimension/longitudinal data/partially linear model/feature screening/sure screening property分类
数理科学引用本文复制引用
郭望,杨孝光,周鹏飞,李运明..超高维纵向数据部分线性模型的特征筛选[J].统计与决策,2024,40(12):46-51,6.基金项目
全军保健专项科研课题(21BJZ39) （21BJZ39）
西部战区总医院军事医学科研课题(2019ZY10 （）
2021-XZYG-A14) （）
中央高校基础研究培育支持计划项目(2682021ZTPY018) （2682021ZTPY018）
