Multi-Material Jetting (MMJ) is an additive manufacturing process empowering the printing of ceramics and hard metals with the highest precision. Given great advantages, it also poses challenges in ensuring the repeatability of part quality due to an inherent broader choice of built strategies. The addition of advanced quality assurance methods can therefore benefit the repeatability of part quality for widespread adoption. In particular, quality defects caused by improperly configured droplet overlap parameterizations, despite droplets themselves being well parameterized, constitute a major challenge for stable process control. This publication deals with the automated classification of the adequacy of process parameterization on green parts based on in-line surface measurements and their processing with machine learning methods, in particular the training of convolutional neural networks. To generate the training data, a demo part structure with eight layers was printed with different overlap settings, scanned, and labeled by process engineers. In particular, models with two convolutional layers and a pooling size of (6, 6) appeared to yield the best accuracies. Models trained only with images of the first layer and without the infill edge obtained validation accuracies of 90%. Consequently, an arbitrary section of the first layer is sufficient to deliver a prediction about the quality of the subsequently printed layers.