Nonsteroidal anti-inflammatory drug solubility optimization through green chemistry solvent: Artificial intelligence technique

Abstract

This research paper presents a comprehensive thermodynamic and heat transfer study on predicting the ternary solubility of Nystatin in SC-CO2-Ethanol (supercritical CO2 and ethanol). The employed process is a thermal-based green processing for preparation of solid nanoparticles. The data collection, consisting of temperature and pressure as input features and ternary solubility as the target variable, was used to train and evaluate four different machine learning algorithms: Random Forest (RF), Extra Trees (ET), NU-SVR, and EPSILON-SVR. The hyper-parameter tuning process employed the Bat Optimization Algorithm (BA), a nature-inspired optimization technique to fine-tune the models and enhance their predictive capabilities. The ET model had a notable R2 score of 0.98526, RMSE of 2.48774E-02, and MAE of 2.13417E-02. The RF model also yielded strong performance, achieving an R2 score of 0.98436, RMSE of 2.55130E-02, and MAE of 2.06314E-02. However, the NU-SVR model exhibited superior performance compared to other models, as evidenced by its remarkable R2 score of 0.99943, thereby showcasing its exceptional precision. The RMSE and MAE for NU-SVR were 4.92372E-03 and 3.94943E-03, respectively, underscoring its precision in predicting ternary solubility. The EPSILON-SVR model, while still respectable, obtained a score of 0.93574 in terms of R2, RMSE of 4.37434E-02, and MAE of 3.79800E-02.

Keywords

• Model prediction
• Machine learning
• Solubility
• Supercritical fluid
• Bat optimization algorithm