Universiti Teknologi Malaysia Institutional Repository

Forecasting crude oil prices using modified empirical wavelet transform with fuzzy C-Means clustering

Md. Khair, Nurull Qurraisya Nadiyya (2019) Forecasting crude oil prices using modified empirical wavelet transform with fuzzy C-Means clustering. Masters thesis, Universiti Teknologi Malaysia.


Official URL: http://dms.library.utm.my:8080/vital/access/manage...


Changes in crude oil spot prices (COSP) have a significant impact on worldwide economy. Therefore, accurate forecasting of COSP is crucial to ensure that necessary steps can be planned earlier by the organizations related to crude oil prices. However, it is difficult to predict accurately the COSP using basic forecasting models because the data are non-stationary and non-linear. Many researchers have empirically proven that the integration of forecasting model with data decomposition method provides superior forecasting results in comparison to basic forecasting model. Nonetheless, most of these hybrid models do not consider the distinction of data characteristics after being decomposed which can affect the forecasting result. In this research, a model called Modified EWT-LSSVM (MEWT-LSSVM) was developed to enhance the forecasting performance of COSP. Empirical wavelet transforms (EWT) was utilized experimentally to separate the nonlinear and time varying components of COSP to address the non-linear and non-stationary issues of COSP. Fuzzy c-means (FCM) clustering was applied to group the decomposed components into several clusters to address the data characteristics issue thus providing better quality inputs for the forecasting model. Each cluster was then forecasted using least square support vector machine (LSSVM), and lastly combined using Inverse EWT to obtain the final forecast. The datasets consisted of daily COSP from West Texas Intermediate (WTI) and European Brent (Brent). For the effectiveness evaluation of the proposed model, the performance of MEWT-LSSVM was compared with EWT-Kmeans-LSSVM, EWT-LSSVM, EWT-Autoregressive Integrated Moving Average (ARIMA), LSSVM and ARIMA models. The experiments produced encouraging results whereby the modified MEWT-LSSVM had 98.87% and 98.86% accuracies for Brent and WTI datasets respectively. Furthermore, comparison of performance between the models demonstrated that the developed model was the most effective for forecasting COSP series to predict accurately oil prices.

Item Type:Thesis (Masters)
Uncontrolled Keywords:fuzzy c-means (FCM), worldwide economy, basic forecasting model
Subjects:Q Science > QA Mathematics > QA75 Electronic computers. Computer science
ID Code:96346
Deposited By: Narimah Nawil
Deposited On:18 Jul 2022 17:53
Last Modified:18 Jul 2022 17:53

Repository Staff Only: item control page