Authors:
Farnod Bahrololloomi
;
Sebastian Sauer
;
Fabio Klonowski
;
Robin Horst
and
Ralf Dörner
Affiliation:
RheinMain University of Applied Sciences, Wiesbaden, Germany
Keyword(s):
Player Modelling, Performance Analysis, Data Science, Computer Games, Electronic Sports (e-sports), League of Legends, Machine Learning, Winning Prediction.
Abstract:
Predicting the outcome of an electronic sports (e-sports) match is a non-trivial task to which different approaches can be applied. While the e-sports domain and particularly the Multiplayer Online Battle Arena (MOBA) genre with League of Legends (LoL) as one of its most successful games is growing tremendously and is professionalizing, in-depth analysis approaches are demanded by the profession. For example, player and match analyses can be utilized for training purposes or winning predictions to foster the match preparation of players. In this paper, we propose two novel performance metrics derived from data of past LoL matches. The first is based on a Machine Learning (ML) based approach and includes individual player variables of a match. The second metric is generally based on heuristics derived from the ML approach. We evaluate the second metric by applying it for winning prediction purposes. Furthermore, we evaluate the importance of different roles of a LoL team to the outcom
e of a match and utilize the findings in the winning prediction. Overall, we show that the influence of a particular role on the match’s outcome is negligible and that the proposed performance metric based winning prediction could predict the outcome of matches with 86% accuracy.
(More)