A Machine Learning Framework for Grain-Storage Deficit Diagnosis in Brazil Using Supervised and Unsupervised Approaches

Authors

  • Daniela Cabral Oliveira Instituto Federal de Educação, Ciência e Tecnologia Goiano(IF Goiano, Campus Rio Verde), Brazil
  • Daniel Emanuel Cabral de Oliveira Instituto Federal de Educação, Ciência e Tecnologia Goiano(IF Goiano, Campus Rio Verde), Brazil https://orcid.org/0000-0002-3824-994X
  • Osvaldo Resende Instituto Federal de Educação, Ciência e Tecnologia Goiano(IF Goiano, Campus Rio Verde), Brazil https://orcid.org/0000-0001-5089-7846
  • Haihani Silva Passos Instituto Federal de Educação, Ciência e Tecnologia Goiano(IF Goiano, Campus Rio Verde), Brazil
  • Deborah de Oliveira Medeiros Freitas Instituto Federal de Educação, Ciência e Tecnologia Goiano(IF Goiano, Campus Rio Verde), Brazil

DOI:

https://doi.org/10.47852/bonviewAIA62028813

Keywords:

grain-storage deficit, machine learning, supervised learning, unsupervised learning

Abstract

This study examines a hybrid approach for diagnosing grain-storage deficits across Brazil by combining supervised classification with unsupervised clustering. Historical records of agricultural production and registered static storage capacity from 2005 to 2021 were obtained from the National Supply Company (CONAB). After cleaning and organizing the records by state and year, the study compared five supervised algorithms—multilayer perceptron, support vector machine with a polynomial kernel, Instance-Based K, J48 decision tree, and Random Forest—with K-means clustering. The outcome was defined before model training: a state-year observation was considered to have a deficit when annual production exceeded registered static storage capacity. This definition is important because the models are reproducing a known operational rule rather than predicting an unknown future event. J48 and Random Forest obtained the highest classification agreement in the reported analysis, reaching 99.79% accuracy. K-means also separated the observations into two groups that corresponded to the predefined classes. The regional analysis showed a persistent difference between production and registered storage capacity in the Brazilian Midwest, where capacity represented approximately 51.8%–54.4% of production in the states examined. The results indicate that the proposed workflow can be useful for organizing historical information, comparing classification methods, and identifying patterns that deserve further investigation. The findings should not, however, be interpreted as evidence of broad forecasting capability. Future studies should include later years and additional information on logistics, climate, markets, storage utilization, and investment. 

 

Received: 16 December 2025 | Revised: 14 July 2026 | Accepted: 9 October 2026

 

Conflicts of Interest 

The authors declare that they have no conflicts of interest to this work. 

 

Data Availability Statement 

The data that support the findings of this study are openly available in the CONAB Agricultural Information Portal at https://portaldeinformacoes.conab.gov.br/. 

 

Author Contribution Statement 

Daniela Cabral Oliveira: Methodology, Software, Validation, Formal analysis, Investigation, Data curation, Writing – original draft. Daniel Emanuel Cabral de Oliveira: Conceptualization, Formal analysis, Investigation, Supervision, Project administration, Funding acquisition. Osvaldo Resende: Conceptualization, Formal analysis, Supervision, Project administration, Funding acquisition. Haihani Silva Passos: Investigation, Data curation, Writing – review & editing. Deborah de Oliveira Medeiros Freitas: Methodology, Software, Validation, Formal analysis, Investigation, Data curation, Writing – review & editing.


Downloads

Published

2026-11-12

Issue

Section

Research Article

How to Cite

Cabral Oliveira, D., Emanuel Cabral de Oliveira, D., Resende, O., Silva Passos, H., & de Oliveira Medeiros Freitas, D. (2026). A Machine Learning Framework for Grain-Storage Deficit Diagnosis in Brazil Using Supervised and Unsupervised Approaches. Artificial Intelligence and Applications. https://doi.org/10.47852/bonviewAIA62028813