Moisture Content Prediction Model for Pharmaceutical Granules Using Machine Learning Techniques

Abstract

Purpose

ML techniques are powerful and novel approaches in modeling fluid bed drying of pharmaceutical granules. The aim of this study was to develop a prediction model and identify relative important factors in the evaluation of moisture content of pharmaceutical granules using ANN and SVM techniques for the datasets of APF.

Methods

ANN and SVM models were developed and compared, utilizing matlab 16.0a as a software tool. Optimizations of the models were also conducted applying GDR and improved TLCO techniques for FFNN and epsilon-SVR, respectively. The performance of the models was evaluated using a quantitative error metric: MAE, MSE, and R².

Results

This study reveals that the FFNN model is an optimal model for predicting moisture content of pharmaceutical granules for the TSG-FBD process model for the datasets of APF.

Conclusions

The model of FFNN, with MSE of 0.0009 and R² of 0.987, is built and accepted as an optimal model for predicting the moisture content of pharmaceutical granules. Temperature, inlet airflow-rate, initial moisture, drying time, and screw speed, respectively are the most important factors in determining the moisture content of the granules.