Accurate prediction of vehicle emissions, particularly carbon dioxide (CO2), is vital for addressing global climate challenges and fostering sustainable automotive practices. This study explores a comprehensive approach to predicting CO2 emissions (g/km) by leveraging traditional machine learning models and a novel hybrid deep learning framework. A dataset encompassing diverse vehicle specifications, including engine size, fuel consumption metrics, and transmission type, was analyzed. Traditional ML models, including linear regression, decision tree, random forest, gradient boosting, and SVM were implemented to establish baseline performance. Among these, the random forest regressor achieved the best performance with an R2 score of 0.93. To further enhance predictive accuracy, a hybrid model was developed, integrating random forest predictions as additional features into a deep neural network (DNN). The hybrid model outperformed all traditional models, achieving an R2 score of 0.95 and lower error metrics, including MAE and RMSE. This approach not only demonstrates the potential of hybrid modeling techniques for improved accuracy but also provides valuable insights into the contribution of key features influencing CO2 emissions. These findings can aid policymakers in formulating effective environmental regulations and support the automotive industry in designing low-emission vehicles for sustainable transportation.