Model selection using a simplex reproduction genetic algorithm. (Q1575168)

In this paper, we show how a simple genetic algorithm (GA) using a simplex reproduction scheme can efficiently be applied to regression model selection. Fast and accurate convergence of this scheme is first demonstrated in an experiment on linear autoregressive model selection and on an extension to nonlinear polynomial models. In a further application, the GA is used for effective center selection of a radial basis function (RBF) network. The center-selection strategy is compared to the well-known orthogonal least-squares learning (OLS) algorithm. For center subsets of various sizes, the GA provides better models than the OLS algorithm.