Extensive data analysis and modelling of carbon dioxide solubility in ionic liquids using chemical structure-based ensemble learning approaches

Abstract

The significant rise in carbon dioxide (CO2) emission due to industrial growth is a major global challenge. As a result, there is a need to implement various techniques to reduce and regulate this phenomenon. One such technique involves the utilization of ionic liquids (ILs) as solvents in CO2 capturing and separation processes, which is already commonly practiced. In this study four advanced intelligent models, Extreme Gradient Boosting (XGBoost), Gradient Boosting (GBoost), Light Gradient Boosting Machine (LightGBM), and Categorical Boosting (CatBoost) have been proposed to predict the solubility of CO2 in 160 different ILs based on factors such as temperature, pressure, and the chemical structure of the ILs. Findings indicate that the XGBoost model is the most accurate among the four models, with the root mean square error (RMSE) and coefficient of determination (R2) values of 0.014 and 0.9967, respectively. Moreover, the results reveal that increasing pressure, decreasing temperature, and lengthening the alkyl chain all increase the solubility of CO2 in ILs. Furthermore, pressure and the number of –CH2 substructure in ILs have the most significant impact on the CO2 solubility in ILs, respectively. To ensure the XGBoost model's reliability, the model's data has been assessed using the leverage technique. The results show that 92.44 % of the data fell within the valid area, which is a substantial percentage and indicates the model's high reliability. The findings of this study will assist in designing and fine-tuning the chemical structure of ionic liquids specifically for CO2 capture purposes.