Artificial neural networks have become a staple in machine learning research and are employed in many interdisciplinary domains. Thus, it is paramount to understand their inner workings when designing and developing new models. One field of research that is particular useful is called visual analytics, which combines interactive visual representations and data analysis algorithms to obtain knowledge. The aim of this thesis is the development of a visual analytics system to analyze the training behavior of a machine learning model predicting microstructure material responses. The goal is to enable the user to explore how different training configurations influence the training process and the model’s performance. In addition, a novel regularization technique and a novel optimization improvement, greedy stochastic permutations, are proposed.