A new decision tree based algorithm for prediction of hydrogen sulfide solubility in various ionic liquids

The solubility of acidic components at various temperatures and pressures in ionic liquids (ILs) is one of the decisive property needed for the appraisal of ILs as potential substitutes for alkanolamines in industrial natural gas sweetening processes, therefore its modeling encompasses scientific and commercial interest. To that end, in the present work, an advanced machine learning approach called stochastic gradient boosting (SGB) tree technique is employed in the calculation of hydrogen sulfide (H₂S) solubility in 11 different ILs within the (303.15 to 363.15) K temperature and (0.0608 to 2.0168) MPa pressure range as a function of critical temperature, critical pressure and acentric factor of ILs accompanied with operational temperature and pressure. A collection of 465 experimental data points were assembled from the literatures. The statistical parameters including correlation coefficient (R) of 0.999543 and mean relative absolute error (MRAE), 0.022198, of the results form dataset values exhibit the high precision of the applied method. Furthermore, the prediction competence of the SGB model has been compared to two well-known equation of states (EOS) as well as Genetic Expression Programming (GEP) and least squares support vector machine (LSSVM) models. According to the results of comparative studies, it was found that the SGB model is more robust, reliable and efficient than other existing techniques for improved analysis and design of natural gas sweetening process.