A1555
Title: Supersaturated design-based statistical methods for variable selection in high-dimensional observational data
Authors: Tharkeshi Dharmaratne - RMIT University (Australia) [presenting]
Alysha De Livera - La Trobe University (Australia)
Stelios Georgiou - RMIT University (Australia)
Stella Stylianou - RMIT University (Australia)
Abstract: In experimental studies, supersaturated designs (SSDs)-based statistical methods are commonly used to screen relevant factors when the number of factors exceeds the run size. Based on simulation studies, several of these SSD-based statistical methods have been shown to perform well in experimental settings. It motivated the exploration of using these SSD methods on high-dimensional observational data for variable selection. Variable selection is a widely used approach for selecting variables of a statistical model in observational studies, which has often been criticized. Therefore, initially reviewed the latest recommendations and methods that are developed for variable selection in observational studies. The performance of the SSD-based statistical methods is then evaluated using both simulated and real-life data, followed by a comparison of their performance with the existing approaches.
