Analysis of CO2 emissions in urban areas using Sumo simulator and machine learning-based prediction methods

The accelerated growth of the vehicle fleet in Latin American cities, such as Quito, combined with its high altitude and high levels of vehicular congestion, has intensified the environmental impact due to carbon dioxide (CO2) emissions. This research proposes a comprehensive methodology that combines realistic traffic simulations using SUMO and machine learning techniques to model and predict CO2 emissions in urban areas. Georeferenced information from OpenStreetMap was used solely to obtain the base map, while vehicular mobility patterns were generated with SUMO based on real data of registered vehicle volumes in Ecuador for the year 2023. Based on these data, mobility scenarios were developed for the Historic Center of Quito, resulting in a substantial dataset that incorporates variations in vehicle density and fleet distribution. Three machine learning models (linear regression, Random Forest, and neural networks) were trained under two experimental configurations. The results show that the Random Forest model, by incorporating both vehicle density and percentage distribution by vehicle type, achieved the best accuracy (R2 = 0.9875 and MAPE = 3.61%). This model was implemented in a web application developed in Streamlit that allows users to input traffic parameters and obtain real-time emission predictions, accompanied by interactive visualizations and export options.

• Autor principal: Casa Vaca, Víctor David (author)
• Formato: bachelorThesis
• Publicado em: 2025
• Assuntos: Emisiones de CO2; Aprendizaje automático; Tráfico vehicular; CO2 emissions; Machine learning; Vehicular traffic
• Acesso em linha: https://repositorio.yachaytech.edu.ec/handle/123456789/1003