Robust composite weighted quantile screening for ultrahigh dimensional discriminant analysis (Q2202032)

The main contribution of this paper is the robust model-free feature screening procedure (robust composite weighted quantile screening, RCWQS) for the ultra-high dimensional discriminant analysis with many categories. The method is validated on both a synthetic dataset and a mixed lineage leukemia real dataset. It performs better than the methods based on mean such as PSIS and VRS, as well as other successful classification screening methods including DC, MV, RRS.