Integrating experimental data and machine learning models for solubility prediction of yellow 23 in supercritical carbon dioxide

Abstract

This study reports, for the first time, the solubility investigation of yellow 23 in supercritical carbon dioxide over a pressure range of 12–30 MPa and a temperature range of 313–343 K. Yellow 23's experimental molar solubilities in supercritical carbon dioxide were found to be between 6.67 × 10⁻⁵ to 20.55 × 10⁻⁵ (313 K), 4.32 × 10⁻⁵ to 23.58 × 10⁻⁵ (323 K), 3.41 × 10⁻⁵ to 27.37 × 10⁻⁵ (333 K) and 2.29.3 × 10⁻⁵ to 3.840.4 × 10⁻⁵ (343 K). Four semiempirical correlations (MST, Chrastil, Bartle et al., and K-J) were used to calculate the solubility of yellow 23 in supercritical carbon dioxide. The machine learning models (Multilayer Perceptron, Gaussian process regression, Random Forest) models were considered for modeling in this research. The K-J model proved to be the most suitable for fitting the experimental data, exhibiting the lowest mean absolute relative deviation of 6.39%. All three machine learning models have impressive act on approximation of yellow 23 solubility. However, model MLP with the highest R² (99.7) and lowest MSE (0.001) was selected as the best among the three models.