Machine Learning-Based CO2 Emission Prediction to Support Sustainable Urban Development

Accurate identification of CO2 emissions from vehicle has become important for sustainable urban planning strategies aimed at mitigating global warming. This study presents a framework for predicting CO2 emissions using data collected from Canada’s complete vehicle fleet. We have applied a voting-based ensemble approach that aggregates five feature selection algorithms such as SelectKBest, Lasso, Recursive Feature Elimination, Random Forest importance, and mutual information to identify the most influential predictors in the dataset. Subsequently, we have evaluated the predictive performance of various machine learning (ML) and deep learning (DL) models using the six highest-ranked features, including combined fuel consumption, engine size, and fuel type. Our analysis shows that the Random Forest Regressor substantially outperforms competing models, achieving a R² value of 0.9976 including the lowest root mean squared error (RMSE) 2.847 g/km. These results highlight the strength of the ensemble framework in generating precise CO2 emission estimates. For sustainable urban transportation planning, the suggested method offers a practical and data-driven framework for decision making. This application can help planners and policymakers to formulate comprehensive strategies to reduce carbon emissions and encourage low-carbon urban development.