Title: Robust clustering for functional data
Authors: Diego Rivera Garcia - Centro de Investigacion en Matematicas (Mexico) [presenting]
Luis Angel Garcia-Escudero - Universidad de Valladolid (Spain)
Agustin Mayo-Iscar - Universidad de Valladolid (Spain)
Abstract: Many algorithms for clustering analysis when the data are curves or functions have been proposed recently. However the presence of contamination in the data can influence the performance of most of these clustering techniques. Therefore, it would be interesting to get available tools for robustifying clustering algorithms. We propose a robust clustering method based on approximate coordinates obtained by applying functional principal components. This robustness is based on the joint application of trimming, for reducing the effect of contaminated observations, and constraints on the variances, for avoiding spurious clusters in the solution. The proposed method was evaluated through a simulation study, which showed an improved performance when compared with other recent methods for functional data clustering.