Quantile-adaptive model-free variable screening for high-dimensional heterogeneous data (Q1952454): Difference between revisions

Analyzing ultra-high-dimensional data with the number of features increasing at an exponential rate is a challenging problem. The first step in this data analysis problem is to use a fast screening procedure to reduce the dimension to a moderate scale. The authors propose a quantile-adaptive model-free screening procedure for this purpose. The approach allows the set of active variables to be different when modelling different conditional quantiles and is effective for analyzing high-dimensional data characterized by heteroscedasticity. The authors use a technique of estimating marginal quantile regression nonparametrically by means of B-spline approximations.