AI-based forecasting of brent crude oil prices: a deep learning and dimensionality reduction approach

Accurate forecasting of crude oil prices is critical for global economic stability, energy security, and informed decision-making by governments, energy companies, and financial institutions. The aim of this study is to improve the accuracy and efficiency of monthly Brent crude oil price forecasts by proposing a novel hybrid DNN-MSE-PCA framework, which integrates deep neural networks with mean squared error (MSE) filtering and principal component analysis for dataset dimensionality reduction. The model is evaluated using multiple accuracy metrics, including mean absolute error (MAE), root MSE, mean absolute percentage error, as well as Diebold-Mariano and Harvey-Leybourne-Newbold tests. Comparative analysis with statistical, machine learning, and deep learning benchmark models demonstrates the superiority of the proposed approach. Empirical results for June–November 2023 show that the MAE decreases from 25.8 $/bbl in DNN-Full model to 12.0 $/bbl in DNN-MSE-PCA, while dataset dimensionality is reduced from 39 to 5 features. The study also highlights the relevance of incorporating geopolitical risk factors, such as the ongoing disruptions in the Strait of Hormuz, which have impacted crude oil price dynamics significantly, moreover the strategic role that crude oil price plays in impacting the ship’s alternatives fuels.