Robust Screening for Ultrahigh Dimensional Data

Variable selection plays an important role in ultrahigh dimensional models. Fan and lv proposed sure independent screening(SIS) based on Pearson correlation. But when outliers exist in data, SIS does not work. In this paper; we propose a new robust method based on composite quantile regression, that is composite quantile regression sure independent screening(CQR-SIS). The method is robust against outliers. Extension simulation studies are conducted to assess the performances of CQR-SIS, SIS, sure independent ranking and screening(SIRS), distance correlation sure independent screening (DC-SIS) and robust rank correlation screening(RRCS). Results show SIS is sensitive to outliers, clearly confirm the effectiveness of the proposed method, that is CQRS is superior to SIS, SIRS, DC-SIS and RRCS.