Controlling the FDR in variable selection via multiple knockoffs

by   Kristen Emery, et al.

Barber and Candes recently introduced a feature selection method called knockoff+ that controls the false discovery rate (FDR) among the selected features in the classical linear regression problem. Knockoff+ uses the competition between the original features and artificially created knockoff features to control the FDR [1]. We generalize Barber and Candes' knockoff construction to generate multiple knockoffs and use those in conjunction with a recently developed general framework for multiple competition-based FDR control [9]. We prove that using our initial multiple-knockoff construction the combined procedure rigorously controls the FDR in the finite sample setting. Because this construction has a somewhat limited utility we introduce a heuristic we call "batching" which significantly improves the power of our multiple-knockoff procedures. Finally, we combine the batched knockoffs with a new context-dependent resampling scheme that replaces the generic resampling scheme used in the general multiple-competition setup. We show using simulations that the resulting "multi-knockoff-select" procedure empirically controls the FDR in the finite setting of the variable selection problem while often delivering substantially more power than knockoff+.



There are no comments yet.


page 37

page 38

page 40


Multiple competition based FDR control

Competition based FDR control has been commonly used for over a decade i...

Competition-based control of the false discovery proportion

Target-decoy competition (TDC) is commonly used in the computational mas...

Normalizing Flows for Knockoff-free Controlled Feature Selection

The goal of controlled feature selection is to discover the features a r...

Variable Selection with the Knockoffs: Composite Null Hypotheses

The Fixed-X knockoff filter is a flexible framework for variable selecti...

Aggregation of Multiple Knockoffs

We develop an extension of the Knockoff Inference procedure, introduced ...

A Generalized Knockoff Procedure for FDR Control in Structural Change Detection

Controlling false discovery rate (FDR) is crucial for variable selection...

Controlling the False Discovery Rate in Structural Sparsity: Split Knockoffs

Controlling the False Discovery Rate (FDR) in a variable selection proce...
This week in AI

Get the week's most popular data science and artificial intelligence research sent straight to your inbox every Saturday.