Predicting results in electronic sports (e-sports) matches is not an easy task. Different methods can be used for this purpose. A well-known video game in the field of Multiplayer Online Battle Arena (MOBA) is the game League of Legends (LoL), which has a relevant professional scene. An important part of professional gaming is analyzing past matches overall and an individual player’s performance to prepare for future matches. In this paper, we follow a design-oriented research methodology (analysis, design, and evaluation) and propose performance metrics that use data from past matches to evaluate a player’s performance. We analyze the necessary data which we acquire by selecting a player, analyzing the player’s latest games, and repeating the process recursively with the players found in his latest games. The data is utilized within a Machine Learning (ML) Model that computes an overall score from individual player variables. From this, we designed a heuristic approach and evaluated it by applying it to the challenge of winning predictions in e-sports. The difference in the influence of the individual player roles on the outcome of the game was also investigated. It was found that this difference is negligible and that the heuristic performance metric can predict the outcome of a game with an accuracy of 86%. Furthermore, the concept of a match calculator is explored, which calculates the outcome of a match using the ML model and different player stats.
Read full abstract