Urban Vehicle Emission Modeling using Machine Learning with OBD/GPS Data and comparison with IVE and EURO 6 Standards

This research focuses on modeling urban vehicle emissions using machine learning techniques applied to OBD and GPS data, aiming to overcome the limitations of traditional models such as IVE and EURO 6 standards. The implemented methodology included collecting emissions, GPS, and OBD data under real-world driving conditions along two Real Driving Emissions routes in Cuenca, Ecuador. A machine learning model was trained using data from the first route, followed by validation using data from the second route. The results were encouraging: the OBD-based model achieved a coefficient of determination (R²) of 0.947 for CO₂, 0.931 for CO, and 0.928 for NO_x, demonstrating high predictive accuracy compared to direct measurements obtained using a PEMS system. Meanwhile, the GPS based model, which estimated dynamic variables such as longitudinal acceleration and road gradient, achieved R² values of 0.916 for CO₂, 0.895 for CO, and 0.891 for NO_x. The emission factors estimated by the OBD model were 171.6g g/km for CO₂, 1.58 g / km for CO, and 0.23g/km for NO_x, with absolute deviations of less than 6% compared to field measurements. The GPS model provided similar emission factors: 176.2g/km for CO₂, 1.71g/km for CO, and 0.26g/km for NO_x further confirming the robustness of both methodologies. These results not only validate the accuracy of the models but also highlight their relevance in estimating emissions in urban contexts, adapting to the specific topography and climate conditions of Cuenca. As such, their use is recommended for developing emission inventories and designing more suitable urban mobility policies, instead of relying on international models such as IVE, which often overestimate emissions under local conditions especially in areas with unique altitudes and driving patterns.