Functional Principal Component Analysis for Extrapolating Multistream Longitudinal Data

The advance of modern sensor technologies enables collection multi-stream longitudinal data where multiple signals from different units are collected in real-time. In this article, we present a non-parametric approach to predict the evolution for an in-service unit through borrowing strength other historical units. Our first decomposes each stream into linear combination eigenfunctions and their corresponding functional principal component (FPC) scores. A Gaussian process prior FPC scores is then established based on semi-metric that measures similarities between streams unit. Finally, empirical Bayesian updating strategy derived update using real-time obtained Experiments synthetic real world show proposed framework outperforms state-of-the-art approaches can effectively account heterogeneity as well achieve high predictive accuracy.