Prediction of winners in MOBA games

Abstract

Multiplayer Online Battle Arena (MOBA) games are very popular in the current eSport scenario, being highlighted in several competitions around the world. However, the domain of knowledge contained in these games is large, which makes it difficult to discover and predict the course of a match. The present work proposes the application of classification algorithms to determine the team with more chances to win a match. Two classifications procedures were used, one based on the composition of heroes in each team and another considering the duration of the match. The experiments were performed on data collected from 123,326 matches of Dota 2, showing that it was possible to achieve approximately 77% accuracy. The results demonstrate the effectiveness of the application when using techniques assisted by computers, and when using the methodology described in championships or other similar games that require the definition of strategies.