Variable Selection via Phony Variables

• Activity Number: 120 - SPEED: Variable Selection and Networks
• Type: Contributed
• Date/Time: Monday, July 31, 2017 : 8:30 AM to 10:20 AM
• Sponsor: Section on Statistical Learning and Data Science
• Author(s): Wenhao Hu* and Eric Laber and Leonard Stefanski
• Companies: North Carolina State University and North Carolina State University and North Carolina State University
• Keywords: Variable selection ; Interpretable ; Error control

Abstract

The performance of penalized methods greatly depends on choice of the tuning parameter. Most existing methods select a single optimal tuning parameter which minimize some criteria, e.g, AIC or BIC. However, it is not intuitive to interpret the tuning parameter selected or the model selected. We propose a new method based on phony variables to choose the tuning parameter. It has asymptotic selection consistency. In addition, it controls error rate for variable selection and outperforms other methods.