Machine learning-based prediction of paracetamol solubility and CO₂ density in supercritical systems using artificial...

Area of Science:

• Chemical Engineering
• Computational Chemistry
• Artificial Intelligence

Background:

• Accurate prediction of solubility in supercritical CO₂ is crucial for pharmaceutical process design.
• Nonlinear interactions between temperature and pressure pose challenges for traditional models.

Purpose of the Study:

• Develop and evaluate an AI-based framework for predicting solvent density and paracetamol mole fraction in supercritical CO₂.
• Compare the performance of Multilayer Perceptron (MLP), Support Vector Regression (SVR), and Tweedie Regression (TDR) models.

Main Methods:

• Utilized temperature and pressure as input variables for AI models.
• Optimized MLP, SVR, and TDR models using the Artificial Rabbits Optimization (ARO) algorithm.
• Performed a comparative analysis of model performance under identical conditions.

Main Results:

• The ARO-optimized MLP model demonstrated superior predictive accuracy for solvent density (R² = 0.99898) and paracetamol mole fraction (R² = 0.96555).
• MLP outperformed SVR and TDR models across all evaluation metrics.
• Nonlinear relationships between solubility, density, temperature, and pressure were identified and visualized.

Conclusions:

• The integrated AI framework with metaheuristic optimization offers a unified approach for supercritical solubility prediction.
• This data-driven methodology provides high predictive accuracy and interpretable insights for pharmaceutical manufacturing optimization.
• The study supports early-stage process optimization in supercritical CO₂ environments.