The presence of invasive cell clusters known as tumor budding and the closely related epithelial mesenchymal transition (EMT) have a prognostic impact on cancer patients' overall survival. Interestingly, data quantitatively analyzing and correlating the amount of tumor buds and patient overall survival as well as the impact of expression of epithelial phenotype markers are missing. Periampullary carcinoma samples of 171 patients were immunohistochemically stained for E-Cadherin (ECad). Tumor cell clusters (TCC, defined from one to 50 cells) were manually quantified comprising tumor cell number and subcellular localization of ECad expression (membranous, cytoplasmic or mixed). Data analyses were performed using elastic net feature selection. Hereby, five distinct intervals of TCC sizes and corresponding fractions of cells with distinct ECad expression were identified. Prognostic features of the defined budding categories were entered into a subsequent Cox regression model together with standard clinicopathological parameters and, based on the model prediction, cases were categorized into "low and high budding" grades. Overall median TCC size was 16 cells (range: 2-36 cells). The median number of TCCs per tumor was 42 (range: 3-283). Elastic net feature selection identified TCCs of 6-10 and 31-35 cells as prognostically most relevant negative and positive features, respectively. Regarding ECad expression, cytoplasmic ECad expression in TCCs of 11-15 as well as of 26-30 cells revealed prognostic relevance. Combining TCC numbers and ECad expression, budding grade qualified as independent prognostic factor for patient overall survival (0.001) in a multivariable clinicopathologic Cox model. Applying an advanced modelling by machine learning on a cohort of periampullary cancers, we show that not the smallest TCCs (1-5 cells) but tumor cell nests containing 6-10 cells display the strongest negative prognostic relevance. Moreover, we demonstrate that larger TCCs might have a strong positive prognostic impact in periampullary adenocarcinomas, contributing to establishing an advanced grading system.