Quantile correlation-based variable selection

Abstract

This article is concerned with identifying important features in high-dimensional data analysis, especially when there are complex relationships among predictors. Without any specification of an actual model, we first introduce a multiple testing procedure based on the quantile correlation to select important predictors in high dimensionality. The quantile-correlation statistic is able to capture a wide range of dependence. A stepwise procedure is studied for further identifying important variables. Moreover, a sure independent screening based on the quantile correlation is developed in handling ultrahigh dimensional data. It is computationally efficient and easy to implement. We establish the theoretical properties under mild conditions. Numerical studies including simulation studies and real data analysis contain supporting evidence that the proposal performs reasonably well in practical settings.