Distribution-free and model-free multivariate feature screening via multivariate rank distance correlation

Feature screening approaches are effective in selecting active features from data with ultrahigh dimensionality and increasing complexity; however, the majority of existing feature either restricted to a univariate response or rely on some distribution model assumptions. In this article, we propose novel sure independence approach based multivariate rank distance correlation (MrDc-SIS). The MrDc-SIS achieves multiple desirable properties such as being distribution-free, completely nonparametric, scale-free, robust for outliers heavy tails, sensitive hidden structures. Moreover, can be used screen responses one dimensional multi-dimensional predictors. We establish asymptotic consistency property under mild condition by lifting previous assumptions about finite moments. Simulation studies demonstrate that outperforms three other closely relevant various settings. also apply multi-omics ovarian carcinoma downloaded Cancer Genome Atlas (TCGA).