Robust feature screening for varying coefficient models via quantile partial correlation

详细信息    查看全文

• 作者：Xiang-Jie Li ; Xue-Jun Ma ; Jing-Xiao Zhang
• 关键词：Quantile partial correlation ; Ultrahigh ; dimensional data ; Feature screening ; Varying coefficient model
• 刊名：Metrika
• 出版年：2017
• 出版时间：January 2017
• 年：2017
• 卷：80
• 期：1
• 页码：17-49
• 全文大小：
• 刊物类别：Mathematics and Statistics
• 刊物主题：Statistics, general; Statistics for Business/Economics/Mathematical Finance/Insurance; Probability Theory and Stochastic Processes; Economic Theory/Quantitative Economics/Mathematical Methods;
• 出版者：Springer Berlin Heidelberg
• ISSN：1435-926X
• 卷排序：80

文摘

This article is concerned with feature screening for varying coefficient models with ultrahigh-dimensional predictors. We propose a new sure independence screening method based on quantile partial correlation (QPC-SIS), which is quite robust against outliers and heavy-tailed distributions. Then we establish the sure screening property for the QPC-SIS, and conduct simulations to examine its finite sample performance. The results of simulation study indicate that the QPC-SIS performs better than other methods like sure independent screening (SIS), sure independent ranking and screening, distance correlation-sure independent screening, conditional correlation sure independence screening and nonparametric independent screening, which shows the validity and rationality of QPC-SIS.