RBF network combined with wavelet denoising for sardine catches forecasting

Abstract

This paper deals with time series of monthly sardines catches in the north area of Chile. The proposed method combines radial basis function neural network (RBFNN) with wavelet denoising algorithm. Wavelet dcnoising is based on stationary wavelet transform with hard thresholding rule and the RBFNN architecture is composed of linear and nonlinear weights, which are estimated by using the separable nonlinear least square method. The performance evaluation of the proposed forecasting model showed that a 93% of the explained variance was captured with a reduced parsimony.