Estimating road transport emissions in Thailand using data-driven modeling approaches

Abstract

Traffic emissions remain a pivotal factor in Bangkok’s air quality issues, particularly as the city expands and vehicle types become more varied and traditional emission-inventory tools, such as Computer Programme to Calculate Emissions from Road Transport (COPERT) and laboratory driving cycles, are useful but often fall short when applied to congested real-world conditions. This work responds to that gap by combining laboratory measurements, regulatory emission factors, and vehicle-registration information to estimate CO2 emissions at the vehicle level. Data from 1,846,111 registered vehicles (January 2023–January 2025) were consolidated from Environmental Protection Agency (EPA) fuel-economy tests, COPERT/Euro standards, and the Thai DLT registry. After harmonizing attributes across all sources, additional features were created and screened using recursive feature elimination (RFE). Three models were trained and evaluated using coefficient of determination (R2), root mean square error (RMSE), and mean absolute error (MAE). CatBoost provided the most reliable predictions (RMSE =23.4 g/km; MAE =14.6 g/km; R2 =0.921), followed by deep neural network (DNN) (R2 =0.887) and support vector regression (SVR) (R2 =0.846). RFE consistently identified fuel consumption measures, emission factors, and tailpipe CO2 as the most influential variables and the results provide a clearer picture of how vehicle technologies contribute to Bangkok’s CO2 burden and support policy measures such as emission taxes, incentives for low-emission vehicles, and monitoring of high-emission groups.