A0273
Title: Efficient stability screening for ultra-high dimensional data
Authors: Ibrahim Joudah - Macquarie University (Australia) [presenting]
Samuel Muller - Macquarie University (Australia)
Houying Zhu - Macquarie University (Australia)
Abstract: Data is ever more complex and high-dimensional in many fields, such as genomics, social science, health, and finance. This presents exciting challenges for statistical analysis. Stability selection, a technique used to stably identify important features, struggles with high dimensionality, often already when the number of features is in their thousands, but even more so when it well exceeds tens of thousands. To address this, stability screening is proposed to screen features stably and efficiently prior to implementing variable selection. Stability screening is a feature screening approach that relies on efficient subsampling techniques, aiming to facilitate stable selection after the initial screening. The latest findings from ongoing research into feasible stability screening are presented. The proposed method for stability screening is illustrated using both simulated and real-world data.