Flexible Structural Varying-Coefficient Regression to Better Predict Outcomes in Complex Neurogenerative Diseases

• Activity Number: 67 - Advances in Variable Selection
• Type: Contributed
• Date/Time: Sunday, August 7, 2022 : 4:00 PM to 5:50 PM
• Sponsor: Section on Statistical Learning and Data Science
• Author(s): Farzaneh Boroumand* and Samuel Muller and Tanya P. Garcia and Rakheon Kim
• Companies: Macquarie University and Macquarie University and University of North Carolina at Chapel Hill and Baylor University
• Keywords: Model selection; High-dimensional data ; Nonlinear interaction; Varying-coefficient regression; Nonparametric approach

Abstract

Brain-related datasets are High-dimensional datasets that have two main characteristics, sparsity and hierarchy. These types of data demand a novel methodology, specifically to capture pre-specified group structures and within-group correlation. The structural varying-coefficient regression (svReg) has been introduced to address these difficulties, however, it cannot detect nonlinear associations and is not robust to misspecification. We will present our most recent work that builds on svReg and that considers such additional challenges. We will present to what extent our proposed models are robust to misspecification and how well they identify nonlinear interactions.