Identifying patterns in combat that are predictive of success in MOBA games

Multiplayer Online Battle Arena (MOBA) games rely primarily on combat to determine the ultimate outcome of the game. Combat in these types of games is highly-dynamic and can be difficult for novice players to learn. Typically, mastery of combat requires that players obtain expert knowledge through practice, which can be difficult to concisely describe. In this paper, we present a data-driven approach for discovering patterns in combat tactics that are common among winning teams in MOBA games. We model combat as a sequence of graphs and extract patterns that predict successful outcomes not just of combat, but of the entire game. To identify those patterns, we attribute features to these graphs using well known graph metrics. These features allow us to describe, in meaningful terms, how different combat tactics contribute to team success. We also present an evaluation of our methodology on the popular MOBA game, DotA 2 (Defense of the Ancients 2). Experiments show that extracted patterns achieve an 80% prediction accuracy when testing on new game logs.