A Deep Learning-Based Ensemble System for Brent and WTI Crude Oil Price Analysis and Prediction

Area of Science:

• Energy Economics
• Computational Finance
• Data Science

Background:

• Accurate crude oil price forecasting is crucial for energy management and investment.
• Deep learning models have shown promise in improving forecast accuracy.

Purpose of the Study:

• To design and evaluate an ensemble prediction system for crude oil prices.
• To compare the ensemble system's performance against individual deep learning models and baseline ensemble methods.

Main Methods:

• Utilized convolutional neural networks (CNNs), long short-term memory networks (LSTMs), bidirectional LSTMs (BiLSTMs), gated recurrent units (GRUs), bidirectional GRUs (BiGRUs), and deep feedforward neural networks (DFFNNs) as individual predictors.
• Employed Bayesian optimization (BO) for fine-tuning model parameters.
• Applied sequential least squares programming (SLSQP) for weighting forecasts in the ensemble stage.
• Implemented extreme gradient boosting (XGBoost) and random forest (RT) as baseline models.

Main Results:

• The proposed ensemble system significantly outperformed individual deep learning models (CNN, LSTM, BiLSTM, GRU, BiGRU, DFFNN).
• The ensemble system also demonstrated superior performance compared to baseline XGBoost and RT models.
• The findings highlight the effectiveness of the proposed ensemble approach in crude oil price prediction.

Conclusions:

• The developed ensemble system offers a more accurate method for crude oil price forecasting.
• This improved forecasting capability can inform energy policy, market supply management, and investment strategies.
• The study provides a robust tool for navigating the volatility of the energy market.