The architecture leverages SUMO (Simulation of Urban Mobility) to generate high-fidelity traffic emission datasets. This data is then processed through a multi-stage ML pipeline to identify spatial-temporal pollution patterns with 85% predictive accuracy.

By applying gradient-boosted trees (XGBoost) and deep neural networks, we developed an optimization algorithm that dynamically redirects traffic flow. Simulations showed a theoretical 75% reduction in CO2 and NOx concentrations at critical intersections.