Ensemble-Based Explainable Machine Learning Framework to Predict CO ₂ Adsorption Capacity of Coal

Integration of carbon capture, utilization, and storage (CCUS) and CO 2 -enhanced CBM recovery (ECBM) serves as a key strategy for mitigating global warming while achieving global net-zero targets. However, precisely predicting gas sorption profiles remains challenging as traditional tests and empirical methods are expensive, time-consuming, and limited by experimental constraints.This study employs ensemble-based ML algorithms, along with explainable AI, to accurately predict the CO 2 adsorption capacity in coal samples across diverse ranks, compositions, and reservoir conditions. The robustness and reliability of the models were validated using 5-fold cross-validation and grid-based hyperparameter tuning. On the basis of the results, the ensemble learning methods were the most competent predictors, with XGBoost performing best ( R 2 = 0.9769, RMSE = 0.064, MAE = 0.033). The developed model could closely describe Langmuir-type adsorption behavior, with good accuracy compared to experimental isotherms. Shapley’s explanations reveal that reservoir pressure and volatile matter exert dominant control on CO 2 adsorption, consistent with adsorption thermodynamics. This rapid, data-driven approach serves as a cost-effective screening-level complement to laboratory adsorption tests, enabling the preliminary evaluation of coal seams for CO 2 -enhanced coalbed methane (ECBM) projects prior to detailed site characterization. By optimizing methane recovery while securely sequestering CO 2, the work advances cleaner coal utilization and contributes to sustainable development and clean energy transitions.