Machine Learning Prediction of CO₂ Emissions from Light-Duty Vehicles in Canada

The accurate prediction of CO₂ emissions from light-duty vehicles is crucial for effective environmental regulation and policy development. Addressing the limitations of previous studies that often rely on single-model approaches, the present research establishes a novel and rigorous performance benchmark by systematically evaluating seven distinct Machine Learning (ML) architectures—ranging from linear baselines to deep neural networks—to identify the optimal predictive framework for the Canadian context. The study evaluated the performance of Linear Regression, Ridge, Random Forest, Gradient Boosting, Support Vector Regression (SVR), K-Nearest Neighbors (KNN), and a Neural Network by utilizing a public dataset of 7,385 vehicles. Following systematic hyperparameter tuning, the Random Forest model demonstrated superior performance, achieving an R² of 0.9982 and a Root Mean Square Error (RMSE) of 2.49 g/km on the test set. Feature importance analysis confirmed that combined fuel consumption is the most dominant predictor of CO₂ emissions. This study establishes a new performance benchmark for CO₂ emission modeling in the Canadian context. The former offers a robust, data-driven tool for regulators and the automotive industry to support emission reduction strategies.