Machine learning analysis and simulation of pharmaceutical drying process based on prediction of concentration distribution and mass transfer

Modeling and analysis of the lyophilization process for low-temperature drying of pharmaceutical compounds was evaluated via a hybrid model that combines mass transfer and machine learning. We investigated the predictive accuracy of three machine learning models- Ridge Regression (RR), Support Vector Regression (SVR), and Decision Tree (DT)-for estimating the concentration (C) in a three-dimensional space, characterized by coordinates X, Y, and Z. Hyper-parameter optimization was performed using the Dragonfly Algorithm (DA) to enhance model performance. Among the models evaluated, the SVR model exhibited superior predictive performance. Excellent generalization was shown by the SVR model's R² test score of 0.999234 and R² train score of 0.999187. The Root Mean Square Error (RMSE) was recorded at 1.2619E-03, and the Mean Absolute Error (MAE) was 7.78946E-04, both reflecting high accuracies. The maximum error observed in predictions was 5.18029E-03, further underscoring the model's precision. These findings show that SVR, when improved with the Dragonfly Algorithm, is a very accurate and dependable way to guess the amount of chemicals in a certain area. This makes SVR a useful tool for chemical engineering and related fields.