Desarrollo de un algoritmo para predecir la producción de maíz en el departamento del Meta a 5 años, utilizando el lenguaje de programación Python.

Abstract

Maize production in the department of Meta shows variations associated with planted area, harvested area, yield, municipality, and production system, which makes medium-term agricultural, logistical, and productive planning more difficult. In this context, this study developed a predictive algorithm in Python to estimate maize production in Meta, using historical microdata from the Rural Agricultural Planning Unit (UPRA) for the 2006–2024 period. The methodology included data collection, cleaning, and organization, as well as the standardization of information by year, municipality, and crop classification. Subsequently, a machine learning model based on Random Forest Regressor was applied, using year, encoded municipality, encoded production technique, planted area, harvested area, and yield as input variables. The model was trained and validated using an 80/20 data split, hyperparameter optimization with GridSearchCV, and performance evaluation through RMSE, MAPE, and R² metrics. The results showed that the proposed algorithm performed better than a traditional three- year moving average method, achieving a 97.91% reduction in RMSE, a MAPE of 7.90%, and an R² coefficient of 0.9943. Likewise, the feature importance analysis identified harvested area and yield as the variables with the greatest influence on production prediction. Finally, a projection was generated for the 2025–2030 period, differentiating between technified and traditional maize. The study concludes that the developed algorithm can serve as a support tool for decision-making in agricultural planning, logistics, input management, storage, and maize supply planning in the department of Meta.