主 题：Strong independence screening for ultra-high dimensional survival data
主办单位：统计研究中心 统计学院 科研处
Ranking by marginal utility provides an efficient way to reduce the data from ultra-high dimension to portable size. In order to handle the complex big data in great variability, the statistic that can measure the nonlinear relationship between response and marginal predictor were extensively discussed recently. Comparing to the regression analysis, it is more challenging when the response is the survival time with possible censoring. We propose a novel method to measure the marginal dependency between survival time and predictors. A screening criteria is presented to determine an active set to include important predictors and exclude unimportant predictors. It is shown that the proposed procedure enjoys good statistical properties. Its performance in finite sample size is evaluated via simulations and illustrated by a real data analysis.