From Data to Decision: Predictive Modeling of Oil Prices Using AutoML and SHAP Analysis

Machine learning and artificial intelligence (ML/AI), once regarded as opaque "black-box" methods, have become increasingly interpretable due to recent advances in explainable AI (XAI). This study proposes an explainable machine learning framework for forecasting crude oil prices by integrating the H2O AutoML platform with SHapley Additive exPlanations (SHAP), thereby achieving both high predictive accuracy and transparent interpretability. Using daily macro-financial data from January 2015 to September 2025 including oil stocks (XLE), the S&P 500 index, industrial production, and the USD index. The study trains and validates a range of ensemble models, with Gradient Boosting Machines (GBMs) emerging as the best-performing models. The results demonstrate strong out-of-sample forecasting accuracy, measured by RMSE in USD per barrel, across different market conditions. Beyond predictive performance, the explainability analysis reveals that oil stocks (XLE), capturing energy-sector equity valuations, exert the strongest positive influence on crude oil prices, highlighting sectoral transmission channels and portfolio rebalancing effects. In contrast, the S&P 500 and industrial production display nonlinear and state-dependent impacts associated with business cycle dynamics, while the USD index exhibits a predominantly negative relationship, consistent with commodity-currency theory. This framework provides a robust approach to oil price forecasting by integrating automated machine learning with interpretable analytics, offering practical insights for investors, risk managers, and policymakers in volatile energy markets.