Time series forecasting based on wavelet decomposition and correlation feature subset selection

• Main Authors: Ahmed, Ehab Ali, Syafiq Fauzi, Kamarulzaman, Gisen, J. I. A., Zuriani, Mustaffa
• Format: Article
• Language: English
• Published: American Scientific Publisher 2018
• Subjects: QA76 Computer software
• Online Access: http://umpir.ump.edu.my/id/eprint/20684/1/40.%20Time%20Series%20Forecasting%20Based%20on%20Wavelet%20Decomposition%20and%20Correlation%20Feature%20Subset%20Selection1.pdf; http://umpir.ump.edu.my/id/eprint/20684/; https://doi.org/10.1166/asl.2018.12976

Due to the possibility of extracting the features of data through wavelet transformation, its use in time series forecasting model has become popular. The appropriate wavelet function selection and the level of decomposition are very necessary for a successful use of the wavelet coupled with the artificial neural network (ANN) models. This is because it can enhance the performance of the model. A drawback of the wavelet-coupled models is their used a large output number to the ANN, thereby making it more difficult to calibrate the neural structure and need a long time to train the model. This study aims to develop a wavelet-coupled ANN for the detection of the dominant input data from the wavelet decomposition sub-series for use as ANN input to increase the model accuracy with minimum input number. The result showed that the Wavelet Transformation and Correlation Feature Subset Selection (CFS) with ANN can significantly improve the efficiency of the ANN models.