Méthode non paramétrique pour l'analyse et la classification des données fonctionnelles

Functional data analysis plays an increasingly important role in many public health and biomedical applications. In particular, such statistical methods provide tools for warping, comparing, averaging, and modeling data involving correlated measurements. In this paper, we present a new approach of regression analysis for classification of functional data. First, we analyze functional observations to capture their key spatio-temporal patterns by searching optimal warping and then estimate the regression function. Next, we investigate different standard representations from literature and estimate the appropriate regression model as a density function. Finally, an example of application involving patients with Rheumatoid Arthritis and healthy subjects as a reference group, is presented.