Exceptional Gestalt Mining: Combining Magic Cards to Make Complex Coalitions Thrive

We propose Exceptional Gestalt Mining (EGM), a variant of Exceptional Model Mining that seeks subgroups of the dataset where a coalition becomes more than the sum of its parts. Suppose a dataset of games in which several roles exist within a team; the team can combine forces from any subset of roles, to achieve a common goal. EGM seeks subgroups for which games played employing a large role set have a higher win rate than games played employing any strict subset of that role set. We illustrate the knowledge EGM can uncover by deploying it on a dataset detailing Magic: The Gathering games: we find combinations of cards that jointly work better in multicolor decks than in decks employing fewer colors. We argue that EGM can be deployed on datasets from sports where several roles exist that directly interact in play, such as ice hockey.