Title: Subspace clustering for functional data
Authors: Yoshikazu Terada - Osaka University; RIKEN (Japan) [presenting]
Michio Yamamoto - Okayama University / RIKEN AIP (Japan)
Abstract: Functional data have the intrinsic high dimensional nature. This nature often makes possible the very good performance in supervised classification for functional data. In the supervised classification problems, it is known that, using the projection into the finite-dimensional subspace, we can extract the intrinsic high dimensional nature from functional data. In the context of unsupervised classification, there are several clustering methods based on the projection into the subspace. However, since these methods mainly focus on within-cluster variance or dimension reduction, the projected data do not necessarily reflect the hidden true cluster structure. A new subspace clustering method for functional data is proposed, which is based on a novel cluster-separation criterion in the finite-dimensional subspace. The proposed method works well not only for the simulated data, but also for the real data which are difficult to obtain a good classification performance by the existing methods.