Computational intelligence investigations on the correlation of pharmaceutical solubility in mixtures of binary solvents: Effect of composition and temperature

Abstract

In this study, we investigated the predictive modeling of solubility as a function of temperature, solvent type, and mass fraction through the implementation of advanced machine learning techniques. The case study is solubility of Rivaroxaban in binary mixtures of dichloromethane and alcohols with varying compositions. Given the non-linear and complex nature of solubility phenomena, we employed three regression models—K-Nearest Neighbors (KNN), Support Vector Machine (SVM), and Matern Gaussian Process Regression (MGPR)—to capture complex interactions between variables. Stochastic Fractal Search (SFS) was utilized for hyper-parameter tuning of the regression models. The model's performance was evaluated using helpful metrics, such as the coefficient of determination (R²), Mean Squared Error (MSE), Mean Absolute Error (MAE), and Mean Absolute Percentage Error (MAPE) to assess accuracy and reliability. Among the models tested, MGPR demonstrated superior performance achieving an exceptional R² of 0.973. This reflects MGPR's strong ability to generalize to new data and capture solubility behavior nuances. In comparison, SVM and KNN models, while moderately accurate, did not match MGPR's precision.