Determination of nanoparticle solubility through green nanonization process using machine learning approach: Computational modeling and optimization

The major aim of the current study is to develop a data-driven methodology based on green processing for estimation of drug solubility in supercritical carbon dioxide as the solvent. Several machine learning algorithms were utilized to simulate Capecitabine solubility in supercritical carbon dioxide for green pharmaceutical manufacturing applications which can enhance the solubility of drugs by this method of processing. In the models, the inputs are pressure (P) and temperature (T), and the target output (Y) is solubility. Tree-based ensemble models of RF (Random Forest), ET (Extra Tree), and GB (Gradient Boosting) were selected for modeling in this research in combination with the optimizer to model the process. The hyper-parameters of models were optimized to reduce the error in the fitting. The coefficient of determination (R² score) values obtained more than 0.96 and RMSE (root mean square error) for ET, GB, and RF models are 2.91, 2.37, and 4.45, respectively. Based on accurate analyses of results Gradient Boosting selected for primary model in this research. The models were able to estimate the drug solubility which can be used to estimate solubility for a wide range, thereby saving time and costs of measurements.