Independent components techniques based on kurtosis for functional data analysis

The motivation for this paper arises from an article written by Peña et al. [40] in 2010,where they propose the eigenvectors associated with the extreme values of a kurtosismatrix as interesting directions to reveal the possible cluster structure of a dataset. In recent years many research papers have proposed generalizations of multivariatetechniques to the functional data case. In this paper we introduce an extension of themultivariate kurtosis for functional data, and we analyze some of its properties. Inparticular, we explore if our proposal preserves some of the properties of the kurtosisprocedures applied to the multivariate case, regarding the identification of outliers andcluster structures. This analysis is conducted considering both theoretical andexperimental properties of our proposal