Functional Prediction of Longitudinal Data

• Activity Number: 439
• Type: Invited
• Date/Time: Wednesday, August 1, 2012 : 8:30 AM to 10:20 AM
• Sponsor: Section on Nonparametric Statistics
• Author(s): Jeng-Min Chiou*+
• Companies: Academia Sinica
• Keywords: Classification ; Functional data ; Prediction ; Subspace projection

Abstract

We present a functional prediction method for longitudinal data that contains heterogeneous patterns of random trajectories. The aim is to predict the future trajectory for an up-to-date and partially observed trajectory. Under the assumption of a mixture of stochastic processes for the sampled trajectories, we develop a probabilistic functional classification method based on subspace projection and a discriminative classification approach. By the clusters defined via subspace projection, we can predict the yet unobserved trajectory by a functional prediction model coupled with the estimated posterior classification probability of the partially observed trajectory. The proposed functional prediction with classification methods are illustrated throughout an application to traffic flow analysis. The methods not only assist in predicting traffic flow trajectories, but also identify distinct patterns of daily traffic flow of typical temporal trends and variabilities.