The serious consequences of global climate change present a significant threat to ecosystems worldwide. The continuous emission of greenhouse gases, mostly carbon dioxide (CO2), has caused a massive rise in world temperatures. A substantial amount of CO2 gas emissions come from transportation, particularly vehicles. Thus, precise quantification of CO2 emissions from vehicles is very important. By doing so, experts can assess the effects of these emissions on the environment and human health in detail. Existing vehicle emission models face challenges due to insufficient, biased, or outdated data on factors like fuel consumption, vehicle activity, and driving conditions, hindering predictive accuracy. Furthermore, these models often operate as black boxes, lacking transparency in the connections between input variables and emissions predictions, as they employ complex algorithms or machine learning techniques without clear interpretation for users. This study conducts a comprehensive analysis evaluating the efficacy of eight machine learning (ML) models for forecasting vehicle CO2 emissions. The techniques employed include Linear Regression (LR), Random Forests (RF), k-Nearest Neighbors (KNN), Support Vector Regression (SVM), Elastic Net (ENet), Extreme Gradient Boosting (XGBoost), Light Gradient Boosting Machine (LGBM), and Bayesian Ridge Regression (BRR). Results indicate that the RF model outperforms all other methods, achieving an impressive R2 score of 99.78%. Additionally, two explainability methods, namely Shapash and ELI5 techniques, are utilized to provide insights into the relevance of input variables and underlying connections influencing emission predictions.