Time Series Prediction using Multiwavelet Transform and Echo State Network

S. M.Abbas, Assad S Abd Alsaada
2013 International Journal of Computer Applications  
The accuracy of forecasts is influenced by both the quality of past data and the method selected to forecast the future. This paper shows a method to accurately predict the time series signal through a combination of decomposition methods and Echo State Network (ESN). Wavelet and Multiwavelet transforms are used to decompose highly nonlinear time series into several stationary time series components. Thereby, they are used to reduce the degree of nonlinear time series and make the issue easy to
more » ... e the issue easy to analyze and predict. These components are fed to an ESN, which predicts the signal. As an illustration for proposed pattern, one of time series signals of the neural network competition (NNC 2010) is used, without knowing its properties. The performances of all the methods used in this work have been evaluated by computer using MATLAB 7.9.0.287 (R2009b) language and RCToolbox version 2.1. Finally, comparison between these two transforms was done in terms of mean square error (MSE). The simulation results showed the effectiveness and significant improvement of the MWT-ESN model compared with DWT-ESN.
doi:10.5120/12207-7663 fatcat:ovnyszttgra3bk75uygtenm4c4