Machine learning-based territorial classification of agricultural profiles in Colombia: A random forest approach
Periodicals of Engineering and Natural Sciences (PEN) · 22.10.2026 · 14 (4) · 153-166
DOI: 10.21533/pen.v14.i4.2103 ↗Kurz gefasst
Fachartikel in der Zeitschrift Periodicals of Engineering and Natural Sciences (PEN), online erschienen am 22.10.2026. Verfasst von Jorge Saul Fandiño Pelayo und 4 weiteren Autor:innen. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).
Abstract (Originalsprache)
Each year, Colombia generates a vast amount of agricultural data reflecting the evolution of production systems across different regions. Municipal Agricultural Assessments (EVA) annually compile records of production, planted areas, harvested areas, and yields for municipalities throughout the country, making them one of the main public sources of information for the agricultural sector. However, these records have primarily been used for compiling statistics and official reports, while their potential to explain differences in agricultural performance across regions using machine learning techniques has been largely unexplored. To leverage this information from a territorial perspective, this research developed a territorial intelligence framework that transforms EVA database records into interpretable agricultural profiles using derived indicators and machine learning models. To this end, five territorial indicators were constructed: real productivity, harvest efficiency, loss index, crop share, and historical variability. These indicators were derived from agricultural records collected between 2019 and 2024 and allowed for the representation of different dimensions of agricultural performance beyond traditional yield measures. The proposed territorial indicators made it possible to distinguish four agricultural profiles describing different patterns of agricultural performance across Colombian agricultural records. Decision Tree and Random Forest models were evaluated for their ability to recognize the proposed agricultural profiles using municipality-level data separation for model validation. Random Forest achieved a classification accuracy of 76.4% ± 0.9%, while Loss Index, Historical Variability, and Crop Participation showed the highest variable importance among the predictors. The resulting agricultural profiles describe different patterns of territorial agricultural performance across Colombian municipalities.
Zitieren
APA
Pelayo, J. S. F., Ravelo, A. M. P., Garzón, J. L. T., Ortega, R. C., & Robles, R. G. (2026). Machine learning-based territorial classification of agricultural profiles in Colombia: A random forest approach. Periodicals of Engineering and Natural Sciences (PEN), 14(4), 153-166. https://doi.org/10.21533/pen.v14.i4.2103
Vancouver
Pelayo JSF, Ravelo AMP, Garzón JLT, Ortega RC, Robles RG. Machine learning-based territorial classification of agricultural profiles in Colombia: A random forest approach. Periodicals of Engineering and Natural Sciences (PEN). 2026;14(4):153-166. doi:10.21533/pen.v14.i4.2103
BibTeX
@article{Pelayo2026,
author = {Pelayo, Jorge Saul Fandiño and Ravelo, Ana Maria Porras and Garzón, Johan Leandro Tellez and Ortega, Rocio Cazes and Robles, Renato Gonzalez},
title = {Machine learning-based territorial classification of agricultural profiles in Colombia: A random forest approach},
journal = {Periodicals of Engineering and Natural Sciences (PEN)},
year = {2026},
volume = {14},
number = {4},
pages = {153--166},
doi = {10.21533/pen.v14.i4.2103}
}
EUVIMED