Genetic algorithms for the selection of smoothing parameters in additive models (Q880886)

A nonparametric additive regression model is considered of the form \[ y_t=\beta_0+\sum_{j=1}^p f_j(x_{ij})+\varepsilon_i, \] where \(y_i\) is the response, \(x_{ij}\) are regressors, and \(f_j\) are unknown regression functions to be estimated via local spline smoothing. A genetic type algorithm is proposed for the selection of a smoothing parameters set for which the smoothed curves minimize the improved AIC criterion. The algorithm uses an improved arithmetical crossover operator described in the paper. The obtained algorithm is compared with Bayesian P-splines, adaptive regression, and some other alternative approaches via simulations.