Cluster analysis with cellwise trimming and applications for the robust clustering of curves

Abstract

In this work, we propose a robust cluster analysis methodology based on cellwise trimming as an extension to a robust version of Principal Component Analysis. This new approach is more reasonable than traditional casewise trimming when the dimension is not small. This type of trimming avoids an unnecessary loss of information when only a few cells of the entirely trimmed observations are atypical. We propose an algorithm to apply this approach. This algorithm is particularized to the case of functional cluster analysis. We provide simulations and applications using real data sets to illustrate the proposed methodology.