Interpretable Monthly Decision Support for Sugarcane Mill Planning: One-Month-Ahead Cane Tonnage Forecasting and Operational-State Profiling in Coastal Ecuador

Monthly cane-supply anticipation is critical for harvest scheduling, mill-intake coordination, transport allocation, labor planning, and maintenance organization in vertically integrated sugarcane agroindustrial systems. This study developed an interpretable monthly decision-support framework using twelve years of original institutional records from Compañía Azucarera Valdez S.A., Milagro, Guayas, coastal Ecuador, covering January 2007 to December 2018. The primary endpoint was one-month-ahead monthly cane tonnage, predicted from variables available at the current monthly time point to avoid look-ahead bias. Complementary diagnostic layers were used to characterize recurrent operational states and to interpret monthly energy-saving behavior. A chronological validation design was applied, with 2007–2015 used for model training and 2016–2018 used for independent testing. The full linear and LASSO models achieved the strongest test performance, with R² values of 0.916 and 0.915 and mean absolute percentage errors near 5%, outperforming random forest and ANN/MLP benchmarks. The area-only baseline was weaker, while the no-area model retained predictive capacity, indicating that harvested area was important but insufficient to explain monthly cane tonnage. PCA and k-means clustering identified four recurrent operational states related to production scale, stress, crop quality, and energy-performance conditions. Monthly energy saving showed ceiling-constrained behavior near 40% and was therefore interpreted as a diagnostic indicator rather than a robust forecasting target. Overall, the framework supports transparent monthly planning by combining leakage-aware forecasting, operational-state interpretation, and conservative energy-performance diagnostics.