Variable selection in function-on-scalar regression

摘要: For regression models with functional responses and scalar predictors, it is common for the number of predictors to be large. Despite this, few methods variable selection exist function-on-scalar models, none account inherent correlation residual curves in such models. By expanding coefficient functions using a B-spline basis, we pose model as multivariate problem. Spline coefficients are grouped within function, group-minimax concave penalty used selection. We adapt techniques from generalized least squares covariance by “pre-whitening” an estimate matrix establish theoretical properties resulting estimator. further develop iterative algorithm that alternately updates spline covariance; simulation results indicate this often performs well pre-whitening true substantially outperforms neglect structure. apply our method two-dimensional planar reaching motions study effects stroke severity on motor control find provides lower prediction errors than competing methods. Copyright © 2016 John Wiley & Sons, Ltd.