Analysis of the impact of highway lane-changing behavior on CO₂ emissions based on NGSIM

To investigate the impact of microscopic driving behavior on traffic carbon emissions, this study extracted lane-changing processes based on the NGSIM dataset and constructed a lane-changing behavior parameter set that includes speed, acceleration, headway, and interaction features with the target lane. The MOVES model was used to calculate the unit-mile CO₂ emissions for each lane-changing process, and an emission prediction model was established using the XGBoost machine learning algorithm. Feature contributions were analyzed using the SHAP method. The results indicate that the three most influential factors on emissions are vehicle average speed, headway standard deviation, and the standard deviation of speed differences with the leading vehicle in the target lane. Based on this, ecological lane-changing strategies are proposed: maintaining speed between 8–14 m/s and controlling fluctuations in headway and speed differences can help reduce highway carbon emissions. The study provides data support and decision-making basis for optimizing green driving behavior.