<p>This paper aims to improve parts production yields manufactured in the binder jetting additive manufacturing process and ultimately facilitate the adoption of mass production automation. This was achieved by investigating the binder jetting printing process variation that was analyzed in sintered part dimensions. The investigation was conducted by designing and performing an iterative experiment where a benchmark stainless steel 316L part duplicates were 3D printed and traced with their specific locations on the build platform. After each print iteration, parts were sintered and their XY dimensions were measured against their nominal CAD. Once all iterations were executed and all parts’ dimensional data were collected, a statistical and a machine learning model were constructed and compared to observe trends and inform predictions. The statistical model was successfully able to inform trends in the dimensional data linearly, but lacked the ability to capture nonlinear relationships between experiment features. Consequently, an initial neural network was developed and trained in order to predict fully sintered parts’ XY dimensions based on their locations on the build platform. The resulting training was unstable with an error as high as 1.4% and the loss converged after 220 epochs. Therefore, the neural network design was refined and robustly trained on the data. The refined design was able to observe consistent trends and led to identify locations that produced varied XY dimensional data. Specifically, the model was successfully able to predict conservatively XY scaling factors that are location unique to 3D print parts within a desired set tolerance. The refined model loss rapidly converged after 6 epochs with only 0.36% error and high prediction accuracy at 99.5%. The process variation was linked to possible hardware inefficiencies, potential varying powder packing density, and particle distribution across the locations of the build platform.</p>