Clustering multivariate functional data using unsupervised binary trees

A model-based clustering algorithm is proposed for a general class of functional data which the components could be curves or images. The random realizations measured with errors at discrete, and possibly random, points in definition domain. idea to build set binary trees by recursive splitting observations. number groups are determined data-driven way. new provides easily interpretable results fast predictions online sets. Results on simulated datasets reveal good performance various complex settings. methodology applied analysis vehicle trajectories German roundabout.