Benzimidazole, a compound valued for its distinct physicochemical characteristics and diverse applications, is the focus of this investigation into its dissolution behavior across various monosolvents. This research employs sophisticated machine learning algorithms, including K-nearest neighbors (KNN), ensemble learning (EL), random forest, decision tree, and adaptive boosting, alongside thermodynamic models such as Apelblat, λh, NRTL, and Margules, to model and predict solubility. A detailed dataset consisting of 171 experimental points, where 136 are used for training and 35 for testing, was curated to encompass 19 monosolvents, including water, methanol, ethanol, n-propanol, isopropanol, n-butanol, 2-butanol, isobutanol, n-pentanol, acetonitrile, acetone, 2-butanone, 1,4-dioxane, methyl acetate, ethyl acetate, propyl acetate, isopropyl acetate, isobutyl acetate, and butyl acetate. The models utilized key input variables including monosolvent type, monosolvent molar mass (g/mol), and temperature (K) which significantly govern Benzimidazole’s solubility. Sensitivity analysis conducted via Monte Carlo simulations identified monosolvent type as the most influential parameter, followed by temperature and monosolvent molar mass, with sensitivity values of 4.28439, 3.54761, and 2.958176, respectively. The dataset underwent thorough validation to ensure its robustness for data-driven modeling. Model performance assessments demonstrated that adaptive boosting achieved superior predictive precision, yielding the highest R² values and the lowest RMSE and AARE percentages across training and test datasets. These findings emphasize the effectiveness of machine learning, especially adaptive boosting, in providing accurate and efficient solubility predictions. The developed machine learning framework provides an economical complement to experimental methods, facilitating rapid and cost-effective prediction of solubility behavior.
扫码关注我们
求助内容：
应助结果提醒方式：
京公网安备 11010802042870号