Determination of Oxaliplatin solubility in SC-CO2: Experimental data, a comparative analysis of sPC-SAFT EoS, semi-empirical correlations and machine learning methods

Abstract

This study presents the first experimental investigation of Oxaliplatin solubility in supercritical carbon dioxide over a pressure range of 120-270 bar and a temperature range of 308-338 K. The experimental data were correlated using the non-cubic equation of state, i.e. sPC-SAFT EoS and six density-based semi-empirical models, including Chrastil, MST, Bartle et al. and Keshmiri et al., and Sodeifian et al. (models I & II). In addition, two machine learning techniques, SVR and GBoost, were applied to model the solubility behavior. Among the semi-empirical correlations, the MST model provided the best agreement with the experimental measurements, yielding the lowest average absolute relative deviation (AARD) of 3.78%. Also, the AARD of sPC-SAFT EoS was 9.34%. Both machine learning approaches demonstrated strong predictive capability; however, the GBoost model outperformed SVR, achieving the highest coefficient of determination (R²=0.996) and the lowest mean squared error (MSE=0.646). Shapley additive explanations (SHAP) analysis identified temperature and pressure as the most influential factors affecting the model predictions. Furthermore, the SHAP results revealed that solubility increases with increasing temperature, pressure, and CO₂ density, which is consistent with established thermodynamic behavior, confirming that the GBoost model successfully captured the underlying physical relationships rather than merely fitting the data.