League of Legends: An Application of Classification Algorithms to Verify the Prediction Importance of Main In-Game Variables

Abstract

League of Legends is currently one of the most popular video games with a competitive scene involving regional and international tournaments. In this paper, we predict the result of competitive matches using information about games played between the years 2016 and 2020 in the most well-known regional and international leagues, such as the Brazilian League of Legends Championship and League of Legends Pro League. We used several different approaches to train our models based on different variable categories. First, we used economic variables registered in the first 10 and 15 minutes of each match. Then we considered variables that change only from the beginning to the end of the game and do not suffer interference before and after the game. The accuracy of classifiers such as KNearest Neighbors, Random Forest, and Decision Tree varied from 68.33% to 85.17%, depending on which variables were used to train the models.