Curves discrimination: a nonparametric functional approach

A new nonparametric tool for studying the relationship between a curve, considered as a functional predictor, and a categorical response is proposed. This is typically a problem of discrimination, also known as supervised classification, but applied to a sample of curves. Starting from a food industry context and a speech recognition problem, we nonparametrically estimate the posterior probability that an incoming curve is of a given class. A consistent kernel estimator is introduced and its practical performance is pointed out by means of a simulation study. Finally, this method is applied to the above-mentioned data sets. (C) 2003 Elsevier B.V. All rights reserved.