Functional Clustering and Identifying Substructures of Longitudinal Data

[[abstract]]This study considers two clustering criteria to achieve different goals of grouping similar curves. These criteria are based on the minimal L2 distance and the maximal functional correlation defined in this study, respectively. Each cluster centers on a subspace spanned by the cluster mean and covariance eigenfunctions of the underlying random functions. Clusters can thus be identified by the subspace projection of curves.[[notice]]補正完畢[[conferencetype]]國際[[conferencedate]]20070625~2007062