Predictive Modeling of CO₂ Solubility in Imidazolium-based Ionic Liquids using Generalized Correlation and Machine LearningAlgorithms

Rising atmospheric CO₂ concentrations have underscored the global demand for efficient carbon capture technologies. Ionic liquids (ILs), particularly imidazolium-based systems, have emerged as promising solvents for CO₂ absorption due to their negligible vapor pressure, high thermal stability, and tunable physicochemical properties. Accurate prediction of CO₂ solubility across diverse IL formulations is essential for optimizing industrial gas separation processes. This study developed a generalized correlation and machine learning (ML) models for predicting CO₂ solubility in disubstituted imidazolium-based ILs paired with tetrafluoroborate ([BF4^-]) and bis(trifluoromethylsulfonyl)imide ([Tf2N^-]) anions. A consolidated dataset of experimental solubility values was used as the basis for model training and evaluation. Extreme Gradient Boosting (XGBoost) and Random Forest (RF) algorithms were systematically developed and benchmarked using temperature, pressure, molecular weight, and anion identity as molecular and thermodynamic input features. Model performance was evaluated using an 70/30 train-test split, 10-fold cross-validation, and statistical metrics including the coefficient of determination (R²), root mean squared error (RMSE), and mean absolute error (MAE). The XGBoost model achieved the highest predictive accuracy, with an R² of 0.9887, RMSE of 0.0247, and MAE of 0.0164, followed by the RF model with an R² of 0.9838. For comparison, the generalized correlation achieved an R² of 0.9602, RMSE of 0.0467, and average absolute relative deviation (AARD) of 10.99%. Beyond predictive accuracy, SHAP analysis indicated that pressure was the most influential variable, followed by temperature, molecular weight, and anion identity. CO₂ solubility increased with pressure and molecular weight but decreased with temperature, while [Tf2N^-]-based ILs exhibited greater CO₂ affinity than [BF4^-]-based ILs, likely due to the larger and more highly fluorinated anion providing greater free volume for CO₂ absorption. Ultimately, this work demonstrates the potential of data-driven approaches to accurately predict CO₂ solubility and support the screening of imidazolium-based ILs for carbon capture applications.