A hybrid modified group method of data handling with the wavelet decomposition for oil palm price forecasting

• Main Author: Basheer, Huma
• Format: Thesis
• Language: English; English; English
• Published: 2021
• Subjects: T Technology (General)
• Online Access: http://eprints.uthm.edu.my/8423/1/24p%20HUMA%20BASHEER.pdf; http://eprints.uthm.edu.my/8423/2/HUMA%20BASHEER%20COPYRIGHT%20DECLARATION.pdf; http://eprints.uthm.edu.my/8423/3/HUMA%20BASHEER%20WATERMARK.pdf; http://eprints.uthm.edu.my/8423/

This thesis presents an exploratory study on hybrid modelling of palm oil price forecasting using modified group method of data handling (GMDH) network with the discrete wavelet transform (DWT) approach. Despite the fact that Malaysia is one of the largest producer and exporter of palm oil, the research on modelling of palm oil price forecasting is still in progress. This study comprises by exploring the appropriate models for forecasting the monthly palm oil price such as conventional GMDH, modified GMDH and hybrid wavelet modified GMDH models. To assess the effectiveness of these models, monthly crude palm oil (CPO) price of Malaysia from January 1983 until November 2019 and Pakistan from September 2001 until June 2019 were used as sample study. The study shows that modified GMDH model, which integrates four transfer functions such as radial basis, sigmoid, tangent and polynomial simultaneously into GMDH, has given the best fit for modelling of palm oil price forecasting as compared to conventional GMDH model. However, the individual model is not best every time to achieve the better results. In improving the model, this study explores a hybrid wavelet modified GMDH model. The architecture of the proposed hybrid model includes DWT, which is selected as a preprocessed clean and pure data enabling the modified GMDH network to present itself as a well-established alternative application to predict the future of CPO. The proposed hybrid model has been applied to different CPO data sets and verified using simulation of different splits of model input data series. Comparative studies among various models were carried out. The mean absolute percentage error (MAPE) of the proposed hybrid model for the monthly CPO price of Malaysia is less than 4 % and coefficient of correlation (R) is 0.99, which show an excellent fit as compared to the individual and other benchmark models. Similarly, the MAPE of hybrid model for Pakistan imports monthly CPO is less than 14 % and R is 0.94, which show good fit as compared to the individual and other benchmark models. The results have demonstrated that the proposed hybrid model is a better alternative model for crude palm oil price forecasting