Pre-manufacturing prediction of jaggery color formation through soil and water parameters using machine learning

Jaggery, a traditional unrefined sugar produced from concentrated sugarcane juice, represents a critical economic sector supporting over 50 million rural livelihoods globally. Jaggery color determines market pricing, with Golden variants commanding 40-60% premium; yet current assessment relies on subjective evaluation and costly chemical interventions. This study presents a machine learning (ML) framework for pre-manufacturing jaggery color classification using agricultural and chemical parameters measurable during sugarcane procurement. Analysis of 2001 sugarcane samples from four varieties across three seasons in Erode, Tamil Nadu, achieved 98.3% test accuracy across three color categories (Golden, Light brown, Dark brown) with Random Forest and Gradient Boosting classifiers. Color labels were assigned through expert visual grading and independently verified under standardized conditions; however, no instrumental colorimetric measurements were used, which constitutes a limitation of the ground truth. Feature importance analysis revealed soil type and water type as the strongest predictive associations within this regional dataset, collectively accounting for 62.93% of predictive importance. Three-tier validation including cross-validation, bootstrap resampling, and Leave-One-Location-Out analysis across 19 locations confirmed robust within-region generalization. Computational efficiency (0.017 s prediction time) supports mobile device deployment. This framework provides quantitative evidence for the predictive associations between cultivation conditions and color outcome, enabling a shift from reactive to proactive quality management. Validation is currently restricted to the Erode region; multi-regional testing is required before broader deployment. No external validation across distinct agroecological systems has been conducted.