Prediction of CO₂ solubility in aqueous and organic solvent systems through machine learning techniques

This study aimed to predict the solubility of carbon dioxide in various organic solvents and water under different temperature and pressure conditions using five machine learning algorithms. A dataset of 1401 samples with eight input variables (six solvents, temperature, and pressure) was analyzed. The green solvents investigated included dimethyl sulfoxide (DMSO), water, ethylene glycol (EG), dimethylformamide (DMF), N-methyl-2-pyrrolidone (NMP), and sulfolane. The results revealed that the random forest and k-nearest neighbor (k-NN) algorithms demonstrated the best performance, achieving zero total error, which indicates their high accuracy in predicting CO₂ solubility. Their strengths include the ability to effectively handle non-linear relationships and robust performance with noisy data. Conversely, the gradient boosting machine (GBM) algorithm exhibited the weakest performance, with a total error of 0.021775, highlighting its sensitivity to noise and complexity in training. Overall, this study underscores the effectiveness of machine learning techniques in predicting carbon dioxide solubility. The insights gained from this research are valuable for enhancing carbon capture and storage processes, as well as improving our understanding of CO₂ behavior in various environmental and industrial contexts. The findings contribute to the ongoing efforts to develop greener and more efficient solutions for managing carbon dioxide emissions.