Open Conference Systems, STATISTICS AND DATA SCIENCE: NEW CHALLENGES, NEW GENERATIONS

Font Size: 
Bootstrap group penalty for high-dimensional regression models
Valentina Mameli, Debora Slanzi, Irene Poli

Last modified: 2017-05-22

Abstract


The paper presents a new penalization procedure for variable selection in
regression models.We propose the Bootstrap Group Penalty (BGP) that extends the
bootstrap version of the LASSO method by taking into account the grouping structure
which may be present or introduced in a model. Based on a simulation study we
demonstrate that the new procedure outperforms some existing group penalization
methods in terms of both prediction accuracy and variable selection quality.