Model-Free Feature Screening for Ultra High-Dimensional Data

• Activity Number: 49
• Type: Invited
• Date/Time: Sunday, August 1, 2010 : 4:00 PM to 5:50 PM
• Sponsor: Section on Statistical Learning and Data Mining
• Author(s): Runze Li*+ and Li-Ping Zhu and Li-Xing Zhu and Lexin Li
• Companies: Penn State and Penn State and Hong Kong Baptist University and North Carolina State University
• Keywords: feature screening ; ultra-high dimensional data ; variable selection

Abstract

Feature screening plays an important in many scientific researches. In this talk, we propose a unified framework of feature screening for various commonly used regression models with ultrahigh dimensional data. We demonstrate that the newly proposed screening procedures possess sure independence screening property without imposing model structure on the regression function. Thus, for a wide range of parametric regression model and semiparametric regression models, the proposed procedures can identify the subset of active predictors with probability approaching to one as the sample size increases. The proposed procedures are computationally efficient in the sense that it does not require to solve a numerical optimization problem with an iterative algorithm. We analyze a real data set and conduct simulation studies to assess the performance of the proposed procedures.