Box-Cox transformation-based annealing robust radial basis function networks for skewness noises

In this paper, a Box-Cox transformation-based annealing robust radial basis function networks (BCT-ARRBFNs) is proposed for training data set with skewness noise. Firstly, the initial structure is determined by a fixed BCT-ARRBFNs model which is derived by support vector regression (SVR). Secondly, the results of the SVR are used as the initial parameters of structure in the fixed BCT-ARRBFNs. At the same time, an annealing robust learning algorithm (ARLA) is used as the learning algorithm for the fixed BCT-ARRBFNs and applied to adjust the parameters and weights. The BCT-ARRBFNs is more generalized radial basis function networks model which has fast convergence speed and is robust against heteroscedasticity noises and outliers. Finally, the proposed algorithm and its efficacy are demonstrated with an illustrative example in comparison with the BCT-ARRBFNs model.