Effective Modeling of CO 2 Emissions for Light-Duty Vehicles: Linear and Non-Linear Models with Feature Selection.

Predictive modeling is important for assessing and reducing energy consumption and CO₂ emissions of light-duty vehicles (LDVs). However, LDV emission datasets have not been fully analyzed, and the rich features of the data pose challenges in prediction. This study aims to conduct a comprehensive analysis of the CO₂ emission data for LDVs and investigate key prediction model characteristics for the data. Vehicle features in the data are analyzed for their correlations and impact on emissions and fuel consumption. Linear and non-linear models with feature selection are assessed for accuracy and consistency in prediction. The main behaviors of the predictive models are analyzed with respect to vehicle data. The results show that the linear models can achieve good prediction performance comparable to that of nonlinear models and provide superior interpretability and reliability. The non-linear generalized additive models exhibit enhanced accuracy but display varying performance with model and parameter choices. The results verify the strong impact of fuel consumption and powertrain attributes on emissions and their substantial influence on the prediction models. The paper uncovers crucial relationships between vehicle features and CO₂ emissions from LDVs. These findings provide insights for model and parameter selections for effective and reliable prediction of vehicle emissions and fuel consumption. [ABSTRACT FROM AUTHOR]