Experimental measurement and machine learning modeling of BBOT solubility in supercritical carbon dioxide

Herein, BBOT solubility in supercritical carbon dioxide was investigated for the first time over a 12–30 MPa pressure range and a 313 to 343 K temperature range. The experimentally determined molar solubility of BBOT in SC-CO 2 was seen to change within the limits of 0.767 × 10⁻ 6 to 2.881 × 10⁻ 6 at 313 K, 0.406 × 10⁻ 6 to 3.286 × 10⁻ 6 at 323 K, 0.256 × 10⁻6 to 3.659 × 10⁻ 6 at 333 K, and 0.207 × 10⁻ 6 to 4.098 × 10⁻ 6 at 343 K. To describe the solubility behavior in mathematical terms, four semi-empirical correlations such as MST, Chrastil, Bartle et al., and K-J were used. Furthermore, two machine learning techniques, namely Gaussian process regression and Support Vector Regression, were employed for the modeling and prediction of solubility trends. The K-J correlation, which was among the semi-empirical models, gave results in the best agreement with experimental data, and the lowest mean absolute relative deviation of 8.308% was produced. However, the Gaussian Process Regression (GPR) model exhibited the highest prediction accuracy among the tested machine learning models, with an R² value of 0.9983 and a mean squared error of 0.002. Therefore, within the studied pressure range of 12–30 MPa and temperature range of 313–343 K, the GPR model provided an accurate representation of BBOT solubility behavior in SC-CO 2 .