A0326
Title: Simultaneous warping and clustering of functional electrocardiogram
Authors: Wei Yang - University of Pennsylvania (United States) [presenting]
Wensheng Guo - University of Pennsylvania (United States)
Abstract: The goal of clustering functional data is to identify distinct functional patterns in the entire domain. These functional data are usually subjected to phase variability distorting the observed patterns and requires curve registration to remove the phase variability. Curve registration requires a target to which a functional object is aligned. A natural target is the cross-sectional mean of the functional objects within the same cluster, which is not available prior to clustering. There is also a trade-off between flexible warping and clustering data into more clusters. The more the phase variability is removed through curve registration, the less the remaining variability in the data, which often leads to a smaller number of clusters. Consequently, the number of clusters based on the amplitude variability and flexibility of the warping are confounded. External information is required to determine the number of clusters and warping flexibility. We proposed an iterative method that performs simultaneous curve registration and clustering. We also proposed a unified criterion for selecting the number of clusters and the penalty parameter of the warping functions. The criterion is derived from the classification likelihood, evaluating the association of the cluster membership with an outcome variable, which penalizes the uncertainties of cluster memberships. We evaluated the method through simulation and applied it to the digital electrocardiographic data.
