This study presents an emission prediction framework for a dual-fuel compression-ignition engine operated on a 20 % winter green oil–diesel blend enriched with hydrogen and equipped with a zeolite-based after-treatment system. Extra Trees, Random Forest, Gradient Boosting, Extreme Gradient Boosting (XGBoost), and AdaBoost are the tree-based ensemble regression models used to predict the emission parameters under limited data conditions. The performance of the models was assessed through 5-fold cross-validation and a 20 % hold-out test method using R-Squared (R²), Root Mean Squared Error (RMSE), Mean Absolute Error (MAE), and Mean Absolute Percentage Error (MAPE) as the evaluation metrics. Among the five tree-based regression models Extra Trees Regressor performed better with highest R² values in the range of 0.99966–0.99974 and the lowest error metrics for all the emission parameters and demonstrates the outstanding robustness and generalization ability of the model. The stronger consistency of extra trees across different test samples was demonstrated by absolute error heatmaps, while the model's accuracy was further validated by comparing actual and predicted values. The study's overall findings demonstrate the potential of tree-based ensemble learning, and extra trees in particular, as a lightweight, accurate and reliable tool for real-time emission prediction in low-carbon dual-fuel systems.