“…large experimental error, low reproducibility, and time and resource consumption. 14 These disadvantages highlight the need for alternative methods to predict powder flowability to save time and resources, especially at the beginning of the development of a new pharmaceutical ingredient, when the amount of material available is at its premium. By compiling and using training data that is the result of this rigorous process of repeat testing, we ensure that the ground truth labels are accurate and hence the predictions made by the Deep Learning (DL) models and the trends which the networks capture are too.…”