Carbon dioxide solubility in polyethylene glycol polymer: an accurate intelligent estimation framework

Polyethylene glycol (PEG), a synthetic polymer made up of repeating ethylene oxide units, is widely recognized for its broad utility and adaptable properties. Precise estimation of CO₂ solubility in PEG plays a vital role in enhancing processes such as supercritical fluid extraction, carbon capture, and polymer modification, where CO₂ serves as a solvent or transport medium. This study focuses on building advanced predictive models using machine-learning approaches, such as random forest (RF), decision tree (DT), adaptive boosting (AdaBoost), k-nearest neighbors (KNN), and ensemble learning (EL) to forecast CO₂ solubility in PEG across a wide range of conditions. The data utilized for model development is sourced from previously published literature, and an outlier detection method is applied beforehand to identify any suspicious data points. Additionally, sensitivity analysis is performed to evaluate the relative influence of each input parameter on the output variable. The results proved that DT model is the most performance method for estimating CO₂ solubility in PEG since it showed largest R-squared (i.e., 0.801 and 0.991 for test and train, respectively) and lowest error metrics (MSE: 0.0009 and AARE%: 22.58 for test datapoints). In addition, it was found that pressure and PEG molar mass directly affects the solubility in contrast to the temperature variable which has an inverse relationship. The developed DT model can be regarded accurate and robust user-friendly tool for estimating CO₂ solubility in PEG without needing experimental workflows which are known to be time-consuming, expensive and tedious.