Extraction of Pharmaceutical Production Factors by Screening of Machine Learning models

Abstract

Improving manufacturing conditions for stable manufacturing is important from the viewpoint of stable supply of pharmaceutical products. The manufacturing process of pharmaceutical products is strictly controlled under GMP, but the quality varies to some extent during continuing the production. If this variance can be reduced, more stable manufacturing becomes possible. The purpose of this study was to extract latent manufacturing factors that lead to a reduction of variance of dissolution ratio, one of the quality parameters of a capsule product, using machine learning. To derive machine learning models, DataRobot, an automated machine learning platform, was used. By the use of multiple models with high prediction accuracy, which were selected by screening of models, we evaluated the influence of manufacturing factors comprehensively from various viewpoints. As a result, "granulation-water temperature" could be extracted as a latent factor. By lowering this temperature, it was estimated that the variance of dissolution ratio reduces.