Dynamic Prediction Intervals for Functional Data

• Activity Number: 595 - Dynamic Methods for Functional Data with Application to Clinical Data Analysis
• Type: Invited
• Date/Time: Thursday, August 3, 2017 : 8:30 AM to 10:20 AM
• Sponsor: ENAR
• Author(s): Nicholas Rios* and Andrada E. Ivanescu
• Companies: Montclair State University and Montclair State University, Department of Mathematical Sciences
• Keywords: functional regression ; dynamic models ; function-on-function regression

Abstract

The prediction of functional data samples has been the focus of several functional data analysis endeavors. We consider the use of dynamic function-on-function regression for dynamic prediction of the future trajectory as well as the construction of dynamic prediction intervals for functional data. The efficacy of Dynamic Penalized Function-on-Function Regression (DPFFR) was assessed and comparisons of DPFFR prediction intervals with other dynamic prediction methods were considered. To make these comparisons, metrics were used to measure prediction error, prediction interval width, and prediction interval coverage. Simulations and applications to biomedical data illustrate the usefulness of the dynamic functional prediction methods.