Functional Variable Selection with Correlated Functional Covariates and Longitudinal Responses

• Activity Number: 38 - Advances in Variable Selection
• Type: Contributed
• Date/Time: Sunday, July 28, 2019 : 2:00 PM to 3:50 PM
• Sponsor: Section on Statistical Learning and Data Science
• Author(s): Rebecca North* and Jonathan Stallrich and Ana-Maria Staicu and Helen Huang and Dustin Crouch
• Companies: NCSU Statistics and North Carolina State University and North Carolina State University and NCSU Biomedical Engineering and University of Tennessee, Knoxville; Mechanical, Aerospace, and Biomedical Engineering
• Keywords: functional data analysis; group lasso; variable selection; electromyography

Abstract

Robotic hand prostheses are capable of translating multiple forearm electromyography (EMG) signals into finger and wrist movement through control strategy. Training the control strategy involves an analysis of concurrent, longitudinal movement and EMG data collected across many forearm muscles, producing highly correlated EMG signals. To improve the prosthetic’s prediction accuracy and stability, we want to identify a control strategy that requires as few EMG signals as possible. We develop a control strategy based on a novel EMG-based functional linear model that accounts for the underlying biomechanics of hand movement, leading to natural, continuous movement of the prosthetic. The model is made parsimonious and interpretable through our proposed Sequential Adaptive Functional Empirical (SAFE) group LASSO procedure motivated by the relaxed LASSO technique. SAFE group LASSO is shown to identify clinically important EMG signals with negligible false positive rates for an able-bodied subject across different postures. A simulation study shows decisive variable selection performance in the presence of correlated covariates and when the covariance structure is misspecified.