Data-driven insights into CO₂ solubility in deep eutectic solvents

Accurate prediction of CO₂ solubility in deep eutectic solvents (DESs) is crucial for advancing carbon capture technologies. This study presents a robust machine learning (ML) framework using key physicochemical properties, including temperature (T), pressure (P), critical temperature (Tc), critical pressure (Pc), critical volume (Vc), acentric factor (ω), and molecular weight (MW), to model CO₂ solubility across 2327 experimental data points derived from 94 unique DESs. Four algorithms, Categorical Boosting (CatBoost), Light Gradient Boosting Machine (LightGBM), Gradient Boosting (GBoost), and Gaussian Process Regression (GPR), were trained and evaluated for this purpose, with CatBoost outperforming other models (R² = 0.998; MAE = 0.021). According to SHAP analysis, pressure and temperature emerged as the most influential parameters, whereas molecular descriptors offered fine-grained adjustments that enriched the predictive performance. The CatBoost model showed high generalizability across diverse conditions and DES combinations, with 94.37 % of predictions falling within the model's valid range. This data-driven approach provides a computationally efficient and interpretable tool for the rapid screening and rational design of high-performance DESs, accelerating the development of advanced carbon capture technologies. • Robust ML framework predicts CO₂ solubility in DESs. • CatBoost model excels with R² of 0.994 and low MAE. • Pressure and temperature are the dominant solubility drivers. • SHAP analysis ensures model interpretability and insights. • High generalizability with 94.37 % of predictions valid.