An Improved FDR Estimate for Variable Selection
Abstract/Description/Artist Statement
Recent work by Luo, Fithian, and Lei (2024) provides a false discovery rate (FDR) estimation procedure for variable selection methods such as LASSO, forward stepwise regression, etc. Their estimator serves as a companion to cross-validation prediction error in assessing the performance of variable selection methods. In this work, we propose an improved version of this estimator through an optimal threshold method that is less biased than the original approach.
Faculty Advisor/Mentor
Dr. Yet Nguyen
Faculty Advisor/Mentor Email
ynguyen@odu.edu
Faculty Advisor/Mentor Department
Department of Mathematics and Statistics
College/School Affiliation
College of Sciences
Student Level Group
Graduate/Professional
Presentation Type
Poster
An Improved FDR Estimate for Variable Selection
Recent work by Luo, Fithian, and Lei (2024) provides a false discovery rate (FDR) estimation procedure for variable selection methods such as LASSO, forward stepwise regression, etc. Their estimator serves as a companion to cross-validation prediction error in assessing the performance of variable selection methods. In this work, we propose an improved version of this estimator through an optimal threshold method that is less biased than the original approach.