Application of neural networks to estimate the emission of polluting gases in relation to the gasoline consumption of light vehicles circulating in Metropolitan Lima, Peru

This study proposes a predictive model to estimate emissions (HC, NOx, CO, and CO2) from light-duty gasoline vehicles in Lima, Peru, based on real-world 2022 consumption data. Unlike previous studies relying on standardized models or external datasets, this work incorporates actual vehicle performance and disaggregates results by body type. Neural networks demonstrated strong generalization, achieving a correlation of R2 =0.85 for CO2, while Random Forest slightly outperformed for HC, NOx, and CO. The proposed framework offers a scalable, data-driven tool to enhance urban emission inventories and support air quality policy in Latin American cities.