Analysis of a nonsteroidal anti inflammatory drug solubility in green solvent via developing robust models based on machine learning technique.

This study develops and evaluates advanced hybrid machine learning models-ADA-ARD (AdaBoost on ARD Regression), ADA-BRR (AdaBoost on Bayesian Ridge Regression), and ADA-GPR (AdaBoost on Gaussian Process Regression)-optimized via the Black Widow Optimization Algorithm (BWOA) to predict the density of supercritical carbon dioxide (SC-CO) and the solubility of niflumic acid, critical for pharmaceutical processes. Using temperature and pressure as input features, ADA-GPR demonstrated the greatest accuracy with R² of 0.98670 (RMSE: 1.36620E + 01, AARD%: 1.32) for SC-CO density and 0.98661 (RMSE: 1.40140E-01, AARD%: 9.14) for niflumic acid solubility, significantly outperforming ADA-ARD (R²: 0.94166, 0.82487) and ADA-BRR (R²: 0.94301, 0.76323). Unlike conventional thermodynamic models, which struggle with generalization across diverse solutes, these models provide robust, scalable predictions over a wide range of conditions. The novel integration of BWOA for hyper-parameter tuning enhances model precision, advancing prior machine learning efforts in supercritical fluid applications. These results establish ADA-GPR as a highly reliable tool for optimizing SC-CO-based processes, offering substantial potential for improving efficiency and sustainability in supercritical fluid-based manufacture for drug processing and other industrial applications.