A Wavelet-Neural Networks Model for Time Series

This work comes as part of the recent continuous and increasing interest in Wavelet Transforms (WT) and Artificial Neural Networks (ANN). This paper introduces a novel hybrid model for solving time series forecast, as a replacement for the classical ARIMA models in order to significantly reduce the error, for it to reach a very small amount closer to zero. The main aim of the paper is to improve the results accuracy by identifying the neural network's input using the (WT), unlike the common classical method used by most researchers to identify the input using the ARIMA model. The results from using both the original method and suggested method for selecting the ANN input were compared using (MSE, RMSE and MAPE) to detect and validate the efficiency of the forecasting. The research concludes that the suggested method shows significant improvement in results over the classical ANN. Error results have dramatically improved, moreover, the improvement measured using MSE approximates to 99.92%. Therefore, showing clearly that the suggested method is both efficient and suitable for processing data in time series forecasting, for the purpose of selecting input for ANN.