Jingxiao Zhang, Xiangjie Li, Haiming Guo

【Abstract】

Variable screening is a very important issue in statistics. In this paper, we propose a new screening, SIS, which do not assume specific models, is robust against outliers. We compare with five methods: Sure Independence Screening, Sure Independent Ranking and Screening, Robust Rank Correlation Screening, Distance Correlation Sure Independence Screening and Martingale Difference Correlation Sure Independence Screening. Simulations indicate that the proposed procedure is significantly better than others.

【Keywords】

ultrahigh dimensional data, robustness, model-free, variable screening