Feature screening filter for high dimensional survival data in the reproducing kernel Hilbert space

We consider the issue of screening uninformative variables associate with survival outcome in ultra high-dimensional situation. specifically, we focus on Hilbert-Schmits Independent criteria (HSIC) statistic defined in the Reproducing kernel Hilbert Space(RKHS) and use the HSIC as a feature screening approach to avoid from difficulty occurred from the ultra high dimensionality. Generally, it is known that the HSIC can account for the flexible relationship between the outcome and predictor variables. Compared with the typical approaches such as regularization and model free screening approaches, our approach does not require the complex non parametric computation or the complicated numerical optimization, Thereby, our proposed approach effectively and flexibly filters out the uninformative predictor. We will show the advantages of our method in numerical simulation studies and illustrate the usefulness of it by applying to diffuse large-B-cell lymphoma data.