Application of Machine Learning for Predicting Bulk Behaviour of Active Pharmaceutical Ingredients

Abstract

The aim of this study was to develop models for predicting powder bulk behaviour from particle properties using machine learning methods. The data consisted of various measurements of particle size, shape, and bulk properties for different active pharmaceutical ingredients. Python libraries were used to pre-process the data, select input features, and train. The models were evaluated using leave-one-out cross-validation and r2 scores. The results showed that the models could predict the flow function coefficient (FFC), bulk density, porosity, and tap density with moderate to high accuracy. However, the models exhibited low prediction accuracy for FT-4 rheometer descriptors. The study demonstrated the feasibility and limitations of using machine learning for powder bulk behaviour prediction.