Variable selection via fixed-X and model-X Knockoff procedures

Mathias Wörndl
2020 unpublished
In many fields of science, a response variable together with a large number of potential explanatory variables are observed and researchers would like to know which variables are truly associated with the response. Variable selection procedures are used to detect such important explanatory variables. FDR-controlling variable selection procedures accomplish this task and make sure that most of the important variables are selected while at the same time, the number of falsely selected variables
more » ... elected variables is not too high. Barber and Candes (2015) introduced such a FDR-controlling variable selection method using so-called Knockoffs to tease apart important and unimportant variables. Using the same ideas, Candes et al. (2018) created a FDR-controlling variable selection method for a setting, where the distribution of the explanatory variables is known and where the number of explanatory variables can be much higher than the number of observations. This thesis presents both techniques in a lecture-note style with detailed proofs.
doi:10.25365/thesis.64231 fatcat:hk63ftywsvbadffhlznpxpkuxm