A novel wavelet artificial neural networks method to predict non-stationary time series

Abstract

There are two main ways to predict a time series: linear and non-linear. But none of them alone is suitable for prediction of data with both linear and non-linear patterns. One of the popular non-linear methods is an artificial neural network. We present a new hybrid prediction method using the artificial neural networks and the wavelet stepwise regression (as a linear method). This method first predicts a time series by the wavelet stepwise regression and then predicts the residuals of this part by the artificial neural network. The performance of the new method is evaluated on simulated and real data sets. The results display that the proposed method predicts the time series better than artificial neural network, wavelet stepwise regression and three other methods.

Keywords:

• Artificial neural networks
• non-decimated wavelet transforms
• stepwise regression
• wavelet filters

Disclosure statement

The authors declare that they have no conflict of interest.

Notes

1 The authors would like to thank Professor. Rob J. Hyndman to provide the real-world data in Time Series Data Library.