Deep learning models to predict CO 2 solubility in imidazolium-based ionic liquids

This study aims to predict CO ₂ solubility in imidazolium-based ionic liquids using deep learning models with input parameters of critical pressure, critical temperature, molecular weight, and acentric factor. The models used in this work include Bayesian Neural Networks (BNN), Deep Neural Networks (DNN), Gradient Boosting Neural Networks (GrowNet), and Tabular Neural Networks (TabNet). The results obtained from this study are compared with two PC-SAFT models named cQC-PC-SAFT-MSA (1) and cQC-PC-SAFT-MSA (2), where deep learning models outperformed SAFT models. Based on graphical and statistical analyses, the GrowNet model, with a root mean square error of 0.0067 and a coefficient of determination of 0.9962, showed the least error compared to other models. In addition, Pearson correlation coefficient (PCC) and Shapley additive description (SHAP) analyses revealed that pressure (P) is a key parameter affecting the solubility of CO ₂ in imidazolium-based ionic liquids and significantly affects the model performance.