Predicting Sugar Yield From Sugarcane Using Machine Learning for Jaggery Production

This study presents a machine learning approach for rapid and accurate prediction of sugar yield from sugarcane per metric ton using multiple physicochemical and environmental parameters. Multiple regression models were developed and evaluated using a comprehensive dataset containing Brix value, Cane juice weight, Pol value, Reducing sugar, Non-sugar content, Water content and type, Soil type, Ratoon type, Season, and Cane variety. After rigorous evaluation and hyperparameter optimization, the Random Forest model demonstrated superior performance with a coefficient of determination (R<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> = 0.9503), Root Mean Squared Error (RMSE = 3.13), and Mean Absolute Error (MAE = 1.63). The feature importance results highlighted a clear hierarchy in predictor significance, withWater content and Brix value collectively driving roughly 90% of the model’s predictive strength. These two variables substantially outweighed all other factors in determining model outcomes. Unlike traditional approaches that rely on proxy measures requiring subsequent calculations and technical interpretation, this direct prediction methodology provides an immediate, actionable metric for all stakeholders. This data-driven alternative to laboratory-based assessment methods enables jaggery manufacturers and sugarcane growers to make informed decisions regarding crop selection, harvesting time, pricing, and processing strategies. The proposed model has the potential to significantly reduce economic losses, improve product quality, and enhance sustainability throughout the sugarcane value chain. Overall, this study offers a simplified, cost-effective, and interpretable solution for sugar yield estimation, which is particularly beneficial to jaggery producers and smallholder farmers who lack access to advanced testing infrastructure.