Predicting Multiplayer Online Battle Arena (MOBA) Game Outcome Based on Hero Draft Data
Weiqi Wang · TRAP@NCI (National College of Ireland) · 2016
DotA 2 is a popular multi-player online battle area (MOBA) game. A critical part of the game play involves choosing from a pool of more than one hundred heroes to form two five-players team. However, as different heroes have their unique attributes and skill sets, selecting a strong combination of heroes (i.e., hero drafting) is an challenging task for new players which requires extensive knowledge and experience. Previous studies have shown that using hero draft data alone can achieve as high as 69.8% of accuracy in predicting game outcomes. However, many aspects in hero draft remains to be further investigated. In this study, we aimed to achieve higher accuracy by adding game length as an input feature. In addition, we used multi-layer feedforward neural networks to predict the game outcome with GPU enabled. However, the results showed that adding game length does not improve the performance significantly nor did neural networks outperform logistic regression significantly.