Modeling and simulation of benzimidazole dissolution behavior in different monosolvents using machine learning and...

Area of Science:

• Physical Chemistry
• Computational Chemistry
• Chemical Engineering

Background:

• Benzimidazole possesses unique physicochemical properties and diverse applications.
• Understanding Benzimidazole's solubility is crucial for its formulation and utilization.
• Accurate solubility prediction aids in optimizing chemical processes and reducing experimental costs.

Purpose of the Study:

• To model and predict the solubility of Benzimidazole in various monosolvents.
• To evaluate the performance of different machine learning algorithms and thermodynamic models for solubility prediction.
• To identify key factors influencing Benzimidazole solubility through sensitivity analysis.

Main Methods:

• Utilized a dataset of 171 experimental solubility points across 19 monosolvents.
• Employed machine learning algorithms: K-nearest neighbors (KNN), ensemble learning (EL), random forest, decision tree, and adaptive boosting.
• Applied thermodynamic models: Apelblat, λh, NRTL, and Margules.
• Performed sensitivity analysis using Monte Carlo simulations.

Main Results:

• Adaptive boosting achieved the highest predictive accuracy (R² values) and lowest error metrics (RMSE, AARE).
• Monosolvent type was identified as the most influential factor affecting solubility, followed by temperature and monosolvent molar mass.
• The machine learning framework provided robust and validated predictions.

Conclusions:

• Machine learning, particularly adaptive boosting, is highly effective for accurate Benzimidazole solubility prediction.
• The developed models offer an economical and efficient complement to experimental solubility determination.
• This approach facilitates rapid prediction of solubility behavior, supporting process development.