CMStatistics 2020: Start Registration
View Submission - CMStatistics
Title: A screening selection method for ultrahigh-dimensional survival data Authors:  Sara Milito - University of Salerno (Italy) [presenting]
Marialuisa Restaino - University of Salerno (Italy)
Francesco Giordano - University of Salerno (Italy)
Abstract: With the recent explosion of computing power, modern studies in many areas, such as medicine, biosciences, demography, economy, generate a large amount of survival data. Selecting significant variables plays a crucial role in model building, and it becomes particularly challenging in an ultra-high dimensional setting where the dimension of covariates can be much larger than the sample size. In this context, since it is crucial to identify the variables that influence the survival time, the main aim is to reduce the dimensionality of the problem. Extensive work has been carried out for variable screening, that is the process of filtering out most of the irrelevant variables. In contrast, all relevant variables survive with probability tending to 1. Most of the methods available in the literature for survival models consider a conditional estimate of the survival function, using the Kaplan and Meier estimator (KM), in order to capture the impact of the variables one by one. This estimator has some disadvantages, especially with continuous covariates, since its formulation does not involve covariates. It is possible to estimate the covariate's effect directly on survival function using a different approach, not involving the KM estimator. Therefore, we aim to suggest a new procedure that overcomes the disadvantages of KM estimator. Some simulation results show that our proposed method performs satisfactorily.