Grouped variables selection via alternating direction method of multipliers in longitudinal analysis. (October 2019)

• Record Type: Journal Article
• Authors: Zhang, Chengyu; Yang, Yihe; Pan, Jianxin
• Is Part Of: Journal of physics. Volume 1224(2019)
• Publication Date: 2019-10
• Subjects: Physics -- Congresses; 530.5
• DOI: 10.1088/1742-6596/1324/1/012104
• Languages: English
• ISSNs: 1742-6588

Abstract

Abstract: This work focuses on automatically and simultaneously selecting and estimating the coefficients of grouped variables in longitudinal data. It is motivated by a large cohort study, the UK North Staffordshire Osteoarthritis Project (NorStOP), which contains highly associated continuous, binary and ordinal variables. To select grouped variables associated with the target response efficiently, we propose a newly-developed generalized estimating equation (GEE) estimation. The innovation of this method is to study and implement GEE with group-lasso rather than traditional lasso which lacks consideration of group effects. We applied the proposed approach to the NorStOP data, and the results turned out to be reasonable.