Explainable Machine Learning Approaches for Climate‑Driven Forecasting of Global Arabica Coffee Prices

• Climate-informed ML framework predicts global monthly Arabica coffee prices. • Random Forest outperformed linear and boosting models in price prediction. • SHAP analysis reveals production and SOI as dominant drivers. • Forecast skill remains robust up to a 3-month lead time. • Explainable AI (SHAP) enhances interpretability for market decisions. Coffee is one of the world’s most important agricultural commodities; yet its market prices are highly volatile due to the combined influence of supply-demand interactions, climate variability and change, and macroeconomic conditions. This study evaluates the predictability of global Arabica coffee prices by integrating climate information, particularly the El Niño Southern Oscillation (ENSO), with advanced machine learning techniques. Monthly production data from leading Arabica coffee-producing countries, Brazil, Colombia, Ethiopia, and Honduras, together with rainfall, temperature, and Southern Oscillation Index (SOI) data for 1990–2023 were used to develop and compare multiple predictive models. Results in model testing phase showed that tree-based models and multi-model ensembles consistently outperformed linear approaches. The Random Forest model achieved the strongest overall performance, with RMSE of 18.67 cents/lb, MAE of 13.47 cents/lb, MAPE of 10.99%, correlation coefficient r of 0.91, and Nash–Sutcliffe efficiency of 0.82. XGBoost and Gradient Boosting also delivered high predictive accuracy, with RMSE of 19.13 cents/lb and 20.78 cents/lb, respectively, and ENS values of 0.81 and 0.78, respectively. Short-term price forecasting at a three-month lead time demonstrated high reliability, whereas predictive skill declined beyond six months. SHAP-based feature attribution analysis indicates that total production and SOI exerted the strongest influence on model outputs, followed by rainfall and temperature, confirming the transmission pathway from climate anomalies to global price fluctuations. These results demonstrate that integrating climate drivers within ensemble machine learning frameworks enhances commodity price predictability and provides a robust, interpretable foundation for climate-informed market intelligence and risk management in coffee-dependent economies.