Applying diverse data-driven paradigms to estimate the density of aqueous solutions of natural amino acid salts

Natural amino acid salt (NAAS) solutions have recently been proposed as efficient solvents for carbon dioxide capture. This research focuses on developing data-driven models to estimate their densities. To this end, a large databank, including 2007 experimental data points of solutions, was collected and then utilized for the models’ development, comparing their performances and selecting the most accurate one. A multiple linear regression analysis identified that the amino acid type, basic compound type, temperature, and mass concentration of the NAAS in the solution are statistically significant enough to model the density of the solutions. The ranking analysis and graphical inspection proved that the least-squares support vector regression (LS-SVR) estimations for the density of these solutions are superior to the other nine models. The developed LS-SVR model estimates 1505 samples of the training stage with the AARD = 0.09 %, MAE = 1.01, RMSE = 2.12, and R-value = 0.9993. Further, 502 testing samples have been predicted by the LS-SVR model with the AARD = 0.31 %, MAE = 3.33, RMSE = 8.48, and R-value = 0.9879. Finally, the leverage method distinguishes 44 potential outliers among the 2007 datasets gathered from the literature.