Exploratory and Predictive Data Analytics of Worldwide Commodity Prices

Abstract

This study examines a diverse dataset consisting of gold, crude oil, corn, natural gas, and nickel commodities. The evaluation encompasses various models, including ARIMA, Random Forest, Deep Neural Network(DNN),Long Short-Term Memory(LSTM),and a hybrid LSTM-DNN model. In-depth time series analysis techniques such as time series decomposition, correlation analysis between different commodities, and stationarity tests are performed. The study also explores different data transformation methods to determine the most suitable approach for each model. Additionally, the research covers different hyperparameter tuning methods such as grid search and the Hyperband algorithm. The performance of each model is assessed and reported using metrics such as Root Mean Squared Error (RMSE), Mean Absolute Error (MAE), and Mean Absolute Percentage Error (MAPE). It also emphasizes the importance of evaluating different data transformation methods, which have not been extensively covered in previous research. By examining and comparing the effectiveness of various models in combination with appropriate data transformation techniques, this study aims to provide valuable information and fill the gap in understanding the impact of data transformation on forecasting accuracy for commodity prices.

Date of Award	Aug 2023
Original language	American English
Supervisor	U Zeyar Aung (Supervisor)