Sampling Studies for Longitudinal Functional Data

We study the data setting consisting of functional data samples repeatedly observed over time. The focus is on the dynamic prediction of the future trajectory. Regression methods based on dynamic functional regression are used for prediction. We propose strategies for the selection of the sampling design for longitudinal functional data. While examining prediction accuracy, it is our aim to identify and recommend 4 months within our historic dataset as candidates for removal. Through this, we will also examine several methods of imputation. Obtaining biomedical data can be costly. We could experience financial relief if we should find that select months in our historic data set can be removed without impacting prediction accuracy. An application to simulated child growth data is presented.