Multivariate integration of time series with ML for corn price forecasting in Colombia

Beltran, Adelaida Ojeda ; Suaza-Medina, Mario E. (Universidad de Zaragoza) ; Zarazaga-Soria, F. Javier (Universidad de Zaragoza) ; De-La-Hoz-Franco, Emiro ; Escorcia-Gutierrez, José

Resumen: The volatility of corn prices poses a significant challenge for both producers and policymakers. This study proposes a hybrid model that combines Extreme Gradient Boosting (XGBoost) and Light Gradient Boosting Machine (LightGBM), optimized through Particle Swarm Optimization with Cuckoo Search (PSO-CS), for accurate corn price forecasting. The approach integrates multivariate time series data, including local prices from the Atlántico market and international futures prices from the Chicago Board of Trade (CBOT). Empirical Mode Decomposition (EMD) is applied to enhance signal clarity and improve model performance. Model performance is assessed through sensitivity analysis and statistical comparison using the Diebold-Mariano (DM) test. The results demonstrate that the proposed ensemble outperforms both individual models and neural network combinations, achieving a Mean Absolute Percentage Error (MAPE) of 2.06.

Idioma: Inglés

DOI: 10.1016/j.eswa.2025.129822

Año: 2026

Publicado en: Expert Systems with Applications 299, Part A (2026), 129822 [23 pp.]
ISSN: 0957-4174

Financiación: info:eu-repo/grantAgreement/ES/DGA/T59-23R

Tipo y forma: Artículo (PostPrint)

Área (Departamento): Área Lenguajes y Sistemas Inf. (Dpto. Informát.Ingenie.Sistms.)