Crude Oil Price Forecasting: A Comparative Analysis of ARIMA, GRU, and LSTM Models

Since 2020, the world hasn’t been able to recover fully from the economic crisis that is caused by the pandemic and the ongoing war between Russia and Ukraine. Crude oil is one of the key ingredients that shapes the world’s economy as of today. This research paper is aimed to predict uncertainty and volatility of crude oil’s prices with time series approaches. To predict crude oil prices, we need to implement Machine Learning and Deep Learning models such as RNN models (LSTM, Bi-LSTM, GRU, and Multi-LSTM) and ARIMA models. To improve the accuracy of the models, we need to preprocess certain data. Picking a variable to be processed is also necessary, in this case we used opening price and date to determine crude oil prices going forward. Data scaling is necessary to maintain numeric data and Ad fuller Test is a must to acknowledge data stationarity. After preprocessing and fitting the data into models, we can see that LSTM and ARIMA stood out within other models, LSTM has an RMSE score of 6.62 and MAE of 4.96 whilst ARIMA has an RMSE score of 6.5 and MAE of 5.03. All the models used are good in predicting crude oil prices, because the R2-Score is ideal, not overfitting and underfitting. This approach on predicting crude oil prices will certainly be impactful in the global economy and avoiding panic bull and bear markets of the crude oil price.