Climatic Intelligence for Coffee: Yield Forecasting in India Using Stochastic Machine Learning and Abiotic Factor Modelling

Accurate forecasting of agricultural yields is vital for enhancing farm-level decision-making, securing food supply chains, and reducing environmental impacts. Coffee, being a globally significant commodity, demands robust predictive frameworks, especially in regions like India, a major producer of Arabica and Robusta varieties. Despite the importance, limited research has examined the role of abiotic and climatological variables in Indian coffee yield prediction, with most existing studies focusing on other coffee-producing countries, thereby restricting region-specific insights. To address this gap, the present study employs long-term datasets (2004–2022) obtained from the Central Coffee Research Institute (CCRI) and Coffee Research Station, Balehonnur, Karnataka. Using stochastic machine learning algorithms, key abiotic factors—including rainfall, temperature, sunshine, humidity, vapor pressure, and dew point—were analyzed through multivariate feature selection and correlation-based grouping. Predictive models such as Bayesian Ridge, Lasso Regression, Elastic Net, Extra Tree, Gradient Boosting, and Random Forest were evaluated. Results demonstrated that Group-3 predictors (Year, Relative Humidity, Rainfall, Temperature) offered the highest accuracy, with Bayesian Ridge and Lasso Regression models achieving R² values of 0.81 and 0.80, respectively, alongside low RMSE values. The findings emphasize precipitation as the most influential variable and highlight the potential of tailored machine learning approaches for reliable, region-specific coffee yield forecasting.