Modelado predictivo de emisiones de CO₂ empleando Machine Learning para la planificación energética sostenible del Ecuador

Predictive modeling of CO₂ emissions using Machine Learning for sustainable energy planning in Ecuador

Abstract

This study develops a machine learning-based predictive model to estimate CO₂ emissions from the Ecuadorian electricity system. A methodology based on the CRISP-ML(Q) approach was used, and hybrid models, including XGBoost, LightGBM, and Random Forest, were evaluated, simulating different energy scenarios between 2026 and 2036. The results show that the optimized Random Forest model presented the best balance between overall fit, stability, and extreme error control (R² = 0.7989), demonstrating adequate capacity to represent the dynamics of emissions in a complex hydrothermal system. Subsequently, different prospective scenarios for 2026–2036 were evaluated, analyzing energy expansion, climate variability, and thermal backup. The results obtained demonstrate that thermal generation and hydrological variability are crucial for determining the dynamics of emissions. These results offer a valuable tool for applications in sustainable energy planning and decision-making within energy transition contexts.