Title: K-means clustering for sparsely sampled longitudinal data
Authors: Michio Yamamoto - Okayama University / RIKEN AIP (Japan) [presenting]
Yoshikazu Terada - Osaka University; RIKEN (Japan)
Abstract: In longitudinal data, the observations often occur at different time points for each subject. In such a case, the ordinary clustering algorithms, such as the K-means clustering, cannot be applied directly. One may apply a smoothing technique to get individual continuous trajectories, followed by finding groups among the trajectories using some clustering algorithm. However, this approach is not appropriate when data of each subject are observed at only a few time points. For sparsely sampled longitudinal data, we develop a new simple clustering algorithm, which can be considered a natural extension of the K-means. We show the consistency of the proposed estimator under mild regularity conditions. Moreover, we investigate the empirical performance of the proposed method through simulation studies and data applications.