Nonparametric feature screening

The measure of correlation between response and predictors plays a critical role in feature ranking and screening for nonparametric regression models. In this paper, a nonparametric function-correlative feature screening is introduced. The newly proposed method does not need any assumption on structural relationships between response and predictors, and among predictors. By using local information flows of model variables, the function-correlation between response and predictors is captured successfully. Selection consistency is achieved as well. Simulation studies are carried out to examine the performance of the new method.