Development of machine learning and integrated design for prediction of pharmaceutical solubility in dense solvent at different temperature levels

This research explores the solubility behavior of Nystatin in a ternary system of supercritical carbon dioxide (SCCO 2 ) and ethanol, focusing on the effects of temperature and pressure. The study employs three distinct machine learning models, Gradient Boosting (GB), K-Nearest Neighbors (KNN), and Polynomial Regression (PR), to predict the ternary solubility of Nystatin in SCCO 2 and ethanol. The hyper-parameter tuning for these models is accomplished using the Dragonfly Algorithm (DA), enhancing their predictive accuracy. The dataset, comprising temperature, pressure, and corresponding ternary solubility values, is analyzed using these models, and their performance is assessed through key evaluation metrics. The GB model achieves a score of 0.863 in R 2 metric. The RMSE and MAE both exhibit values below 0.1, thereby indicating the model’s high level of precision. The K-nearest Neighbors (KNN) algorithm demonstrated a high level of accuracy, as evidenced by its impressive coefficient of determination (R 2 ) of 0.978. This result suggests that the KNN model exhibited a strong alignment with the underlying data. The performance of the PR model is outstanding, by its high R 2 score of 0.991, indicating its precise representation of solubility trends. The low RMSE and MAE values further confirm its precision.