摘要
变量筛选是分析超高维数据的重要方法,现有边际筛选方法在某些情况下易导致高错误发现率。另外,在很多实际问题中,研究者事先知道某个(些)协变量对响应变量有重要影响,在变量筛选的过程中,应该考虑这个先验信息。文章提出了一种新的条件变量筛选方法,并证明了该方法具有确定性筛选性质。数值模拟结果验证了所提方法在有限样本下具有良好表现。最后,通过实际数据分析验证了该方法的有效性。
Variable screening is an important method to analyze ultrahigh dimensional data.The existing marginal screening methods tend to result in high false discovery rate in some cases.Additionally,in many practical problems,the researchers have prior knowledge of a certain covariable(or some covariables),which has significant effect on response variables.During the variable screening process,the prior information should be considered.This paper proposes a new conditional variable screening approach,and proves its deterministic screening property.Numerical simulation results show that the proposed method performs well in limited samples.Finally,the paper verifies the validity of this method by actual data analysis.
作者
刘伟
曹智苗
陈晓林
Liu Wei;Cao Zhimiao;Chen Xiaolin(Chinese Academy of Education Big Data,Qufu Normal University,Qufu Shandong 273165,China;School of Statistics,Qufu Normal University,Qufu Shandong 273165,China)
出处
《统计与决策》
CSSCI
北大核心
2021年第5期30-33,共4页
Statistics & Decision
基金
山东省自然科学基金面上项目(ZR2020MA023)
关键词
超高维数据
变量筛选
条件变量筛选
条件累积分布函数
ultrahigh dimensional data
variable screening
conditional variable screening
conditional cumulative distribution function
