Machine learning models for predictive accuracy and causal interpretation of sugarcane yield

Honorio M.A.; Rodríguez H.I. · CISCI - Conferencia Iberoamericana en Sistemas, Cibernetica e Informatica (2025) · DOI: 10.54808/cisci2025.01.152

Resumen

The study tested the predictive and explanatory power of irrigation water and nitrogen applied per hectare, number of harvest cuts, and temperature on sugar yield per hectare using machine‑learning models. We trained artificial neural networks, support vector regression, and random forests, tuning hyperparameters with Bayesian optimization and cross‑validation. Data came from 240 H‑32 sugarcane plots in Ascope Province, Peru. Random forests produced the lowest prediction error. SHAP (SHapley Additive exPlanations) confirmed its interpretability, showing that the number of cuts was the most important variable for explaining sugar yield per hectare.

Instituciones: Universidad Privada del Norte