Predicting time series with wavelet packet neural networks

Inspired by both the multilayer perceptron (MLP) and wavelet decomposition, Zhang and Benveniste (1992) proposed the wavelet MLP (W-MLP), which has been used for time series prediction. The wavelet packet MLP (WP-MLP) is an MLP with the wavelet packet as a feature extraction method to obtain time-frequency information. The WP-MLP has been successfully applied to biomedical, image and speech classification. In this paper, we demonstrate that the WP-MLP can be a very promising approach to time series prediction.