Using proposed hybrid method for neural networks and wavelet to estimate time series model

Abstract EN

The research presents a new hybrid model that proposes its use for accurate time series prediction, which combines wavelet transformations to remove de-noise of the data before using it in artificial neural network and applied for time series.

To find out the effectiveness and efficiency of the proposed method on artificial neural network models in prediction, the proposed method was firstly applied to the generation time series data (first-order auto-regression) through several simulation examples by changing the value of the parameters and sample size with the generation data being repeated 25 times, secondly the application on the real data represents the monthly average of the price of an ounce of gold in the Kurdistan Region, To compare the simulation results and the real data of the proposed and traditional method, then design a program in Matlab language for this purpose and based on the criteria (MSE, MAD, R2).

The results of the research concluded that the proposed method is more accurate than the traditional method in estimating the parameters of the time series model.