Explainable and Robust Oil Price Forecasting Using Hybrid Deep Learning and SHAP Interpretability: A Long-Term Analysis of WTI Prices from 1986 to 2025

Creative Commons License

Polat O., Parlak N., Söğüt E., Fendoğlu E., Türkoğlu M.

Gazi Mühendislik Bilimleri Dergisi, cilt.12, sa.2, ss.201-220, 2026 (TRDizin)

Özet

Crude oil is an essential input for strategic sectors such as petrochemicals, industrial production,transportation, defense, and manufacturing. Fluctuations in crude oil prices affect not only energy marketsbut also global financial stability and policymaking. Therefore, accurate crude oil price forecasting iscrucial for maintaining energy market stability, supporting sustainable energy policies, and improvingeconomic planning. This study proposes a hybrid deep learning model that combines ConvolutionalNeural Networks (CNN), Bidirectional Long Short-Term Memory (BiLSTM), and Multi-Head Attentionmechanisms for crude oil price forecasting. The proposed architecture uses multi-scale convolutionallayers to capture temporal patterns at different scales. BiLSTM layers learn sequential dependencies, whilethe attention mechanism highlights the most informative features. The dataset includes West TexasIntermediate crude oil spot prices and macroeconomic and energy market indicators from January 1986 toMarch 2025. This period covers approximately 39 years of market history. The proposed model achievedan RMSE of 0.1994, an MAE of 0.1641, and an R² score of 0.9224 on the test set. Model robustness wasfurther evaluated under different market regimes, including the COVID-19 shock, the recovery period,and the post-pandemic market environment. SHapley Additive Explanations (SHAP) analysis was alsoperformed to identify the most influential predictors. The findings indicate that the Consumer PriceIndex, jet fuel prices, gasoline prices, diesel prices, and other energy-related variables are among the mostimportant predictors of crude oil prices. Overall, the proposed framework effectively captures thenonlinear behavior of crude oil markets while providing transparent and interpretable predictions.