文章基本信息

标题：Adaptive model-free sure independence screening
本地全文：下载
作者：Canhong Wen ; Shan Zhu ; Xin Chen 等
期刊名称：Statistics and Its Interface
印刷版ISSN：1938-7989
电子版ISSN：1938-7997
出版年度：2017
卷号：10
期号：3
页码：399-406
DOI：10.4310/SII.2017.v10.n3.a4
出版社：International Press
摘要：Variable screening procedure is popularly used in ultrahigh-dimensional data analysis. It ranks the importance of the predictor variables by marginal correlations and then screens out the variables that are weakly correlated or uncorrelated with the response variables. Though demonstrated their effectiveness, the performance of most variable screening approaches depend on the pre-determined threshold of the size of selected predictor variables, which is some integer multiples of $\lceil n / \log(n) \rceil$ with $n$ being the sample size. To circumvent this issue, we propose a novel data-driven variable screening procedure that can automatically determine the threshold. In our proposal, we rank the importance of the predictor variables by the $p$-values using some modified independent tests, with the smaller $p$-values indicating higher correlation. Compared with the existing counterpart, extensive simulation studies and a real genetic data indicate the preference of our procedure.
关键词：adaptive threshold; distance correlation; false discovery rate; sure independence screening; ultra-high dimensional data