Classification methods for multivariate functional data with applications to biomedical signal

In this work we propose a multivariate functional clustering technique based on a distance which generalize Mahalanobis distance to functional data generated by stochastic processes. This new mathematical tool is well defined in L2(I), where I is a compact interval of R, and considers all the infinite components of data basis expansion while keeping the same ideas on which Mahalanobis distance is based. To test the robustness of our clustering procedure we first present some simulations, comparing the performances obtained using our distance and other known distances, eventually applying it to a dataset of reconstructed and registered ECGs.