Prediction of the solubility of organic compounds in high-temperature water using machine learning

The estimation of the solubility of organic compounds in high-temperature water is important for designing chemical processes. This study aimed at predicting the solubility of organic compounds in high-temperature water in the range of 100–250 °C using machine learning. The chemical structure of the organic compound was converted into 196 descriptors (parameters) using an open-source toolkit. The experimental solubility data were regressed using the descriptors, temperature, and water density. The regression methods of ordinary least squares, least absolute shrinkage and selection operator (Lasso), and support vector regression (SVR) were compared. A regression method combining the Lasso and SVR (Lasso + SVR) was developed. The model thus obtained this method was found to accurately predict the solubility of organic compounds in high-temperature water, with a root-mean-square error of 0.5. The findings in this study would be useful for predicting the solubility of any organic compound in high-temperature water. [Display omitted] •Method proposed for solubility prediction of organic compounds.•Predictions based on chemical structure, temperature, and water density.•Lasso + SVR method shows improved predicting performance.•Preselection of parameters using Lasso improves SVR performance.