B1634
Title: A powerful and precise filter of feature selection using group knockoffs
Authors: Jiaqi Gu - Stanford University (United States) [presenting]
Zihuai He - Stanford University (United States)
Abstract: Selecting important features that have substantial effects on the response with provable type-I error rate control is a fundamental concern in statistics with wide-ranging practical applications. Existing knockoff filters, although shown to provide a theoretical guarantee on false discovery rate (FDR) control, often struggle to strike a balance between high power and precision in pinpointing important features when there exist large groups of strongly correlated features. To address this challenge, a new filter is developed using group knockoffs to achieve high power and precision in pinpointing important features. By additionally taking the group structure of features into consideration, the proposed filter is proven to provide valid control on FDR. With detailed procedures for both powerful but intensive exact inference and computationally efficient surrogate inference, the proposed filter is evaluated in extensive simulations and is applied to a real Alzheimer's disease genetics dataset. Via experiments, it is found that the proposed filter can not only control the proportion of false discoveries but also pinpoint the most important features precisely.
