New Study Explores Forecasting Robusta and Arabica Coffee Prices 2 to 6 Months in Advance Through Machine Learning – CoffeeTalk

Machine learning techniques were utilized to forecast Arabica and Robusta coffee prices 2 to 6 months in advance, employing a dataset comprising climatic, production, and economic indicators from 2009 to 2025. Utilizing SHAP analysis, crucial factors influencing price variations were identified, including global and Brazilian stock levels, the impact of Vietnamese drought, Colombian rainfall, and fluctuations in the U.S. dollar exchange rate. Notably, even variables with minimal direct correlation exhibited significant predictive capabilities.

The study compared various machine-learning models: Multilayer Perceptron (MLP), Extreme Learning Machine (ELM), Random Forest (RF), XGBoost, and combinations of these models through stacking configurations. For Robusta, the stacking configuration of RF/XGBoost/ELM achieved the highest accuracy, marked by a Pearson correlation coefficient (PA) of 0.9264, a Hit Rate (HR) of 92.80%, a Root-Mean-Square Error (RMSE) of 0.0793, and a Mean Absolute Percentage Error (MAPE) of 9.3051%. In contrast, forecasting Arabica was more challenging, with ELM achieving the highest independent test performance, reflected in a PA of 0.9064 and an HR of 94.1167%.

Assessing model robustness against future climate scenarios revealed that, despite stringent conditions predicting higher temperatures and reduced production, the models did not forecast significant price surges. Forecasts for Arabica proved more difficult, while predictions for Robusta aligned better with realistic market expectations.

The study emphasizes the efficiency of machine-learning models and SHAP interpretation in unveiling the complex interplay of economic, climatic, and productive factors affecting global coffee prices. It highlights the conservative nature of the models in extrapolating future prices under unprecedented climatic changes. Understanding these dynamics is essential, as coffee remains a vital agricultural commodity, significantly impacting global economic frameworks and the livelihoods of millions.

The global coffee market is primarily dominated by Arabica (56.6%) and Robusta (43.4%) production, with Brazil and Vietnam being the leading producers. Coffee’s inherent demand structure is characterized by price volatility due to varied climatic influences, production variations, stock policies, and macroeconomic uncertainties. Given coffee’s sensitivity to climatic conditions, effective forecasting becomes crucial for stakeholders.

Machine learning emerges as a suitable solution to integrate diverse predictors and reveal complex interconnections within the market, unlike traditional time-series models like ARIMA which may fail to capture intricate non-linear relationships. This study fills a notable gap by leveraging advanced modeling techniques to enhance predictive accuracy, addressing both ecological and economic challenges inherent to the coffee industry. It aims to deliver significant insights for producers, traders, and policymakers to navigate market volatility and strengthen the resilience of the global coffee value chain.

Read More @ Nature