College of Computer Science and Engineering
Information & Computer Science Department
Presents Public Seminar
Forecasting Price of Energy Commodities Using Individual Machine Learning Models and Homogeneous Ensemble Models
Date: 17th February, 2021
Time: 03:30 pm – 4:15 pm
Location: MS Teams Link: Click here to join the meeting
Mr. Kazi Ekramul Hoque
At present, the essential energy resources are crude oil, natural gas, and coal because these are the highest consumed energy sources in the world. As a result, the price of these energy commodities plays a vital role in the world economy. Forecasting these energy commodity prices can enable authorities to take the necessary steps to minimize economic depression and enhance economic advancement. Moreover, building a forecasting model for crude oil prices with reasonable accuracy is challenging due to irregular oil price fluctuations. We proposed different machine learning models to forecast the price of crude oil, natural gas, and coal and compare our proposed models with the existing models using a dataset provided by the International Monetary Fund (IMF). We performed time series forecasting by developing the proposed individual and ensemble machine learning models to forecast these energy commodities' prices. Then we performed numerous sets of experiments by running the model using time split cross-validation to get the best hyper parameters of these proposed models. Finally, we used the tuned models to forecast the price of crude oil, natural gas, and coal in the test set. We evaluated the performance of the proposed models using root mean square error and mean absolute percentage error. Two of our proposed models XGBoost and Support Vector Regression, surpassed all other models by scoring remarkably low root mean square error (RMSE) 5.32 and 5.65, respectively, in the test set in forecasting crude oil price. Furthermore, a statistically significant difference from other machine learning models is observed with a confidence level of 95% using the Wilcoxon statistical test. Our proposed models can assist the appropriate authority in forecasting these energy commodity prices early and precisely. Moreover, our study will pave the way to develop more accurate price forecasting models.
Mr. Hoque is a MSc Student in Computer Science in the ICS Department at KFUPM. His primary research interests involves time series forecasting, machine learning, and ensemble learning
All faculty, researchers and graduate students are invited to attend.
Information & Computer Science Department, College of Computer Sciences and Engineering
Telephone: +966 (13) 860 2175, Email: firstname.lastname@example.org, Website: www.kfupm.edu.sa/departments/ics/
Copyright © 2014 King Fahd University of Petroleum & Minerals