Local smoothing regression with functional data (Q964614)

A kernel-type estimate is considered for the estimation of a nonlinear regression operator \(r(x)=E(Y\,|\,X=x)\), where \(Y\in R\), and \(X\) is a random element in some semi-metric functional space. A local bandwidth selection rule is proposed based on a weighted cross-validation criterion. It is shown that this choice of bandwidth is asymptotically optimal. Results of simulations and application to spectrometric curves are presented.