Modeling of CO2 absorption capabilities of amino acid salt solutions using intelligent paradigms: A statistical and feature importance analysis

In this paper, modeling of carbon dioxide (CO 2 ) absorption by various amino acid salt solutions is presented as a function of operational parameters of different machine learning algorithms, including extreme gradient boosting (XGBoost), extreme learning machine (ELM), adaptive boosting support vector regression (AdaBoost-SVR), and multivariate adaptive regression spline (MARS). A dataset of 2351 data on carbon dioxide absorption by 17 different amino acid salt (potassium- and sodium-based) solutions was collected from the literature at different molecular weights, temperatures, partial pressures, and concentrations. To analyze the suggested model performances, different statistical criteria were employed. XGBoost followed ELM exhibited superior performance in forecasting CO 2 absorption capabilities of amino acid salt solutions with RMSE, R 2 , SD, and AARPE of 0.0314, 0.997, 0.072044, and 3.3890, respectively. Moreover, the models' performances were clarified by visually comparing their outputs with experimental CO 2 absorption data. Through comparative analysis, it has been determined that the XGBoost model exhibits the highest level of accuracy in estimating CO 2 absorption. Along with the black-box models, a white-box regression correlation derived from genetic programming (GP) was also developed to provide transparency and practical value for modeling and simulation tasks. Sensitivity analysis indicated that P CO2 and concentration had the most significant direct and inverse effects on CO 2 absorption process with a relative importance of 0.4276 % and − 0.2267 %, respectively. Feature importance values based on permutation and SHAP methods along with the corresponding Mutual Information coefficients indicate the P CO2 as the most informative and important input feature. • Different ML methods were compared to estimate CO 2 solubility in different amino acids. • 2351 data related to CO 2 absorption in 17 amino acid salt in various practical conditions were used. • Molecular weights, temperatures, partial pressures, and concentrations were inputs. • XGBoost model was found as the most efficient and reliable model. • SHAP method was applied to predict the influence of each input feature.