A Comparison of Machine Learning Algorithms for Predicting Winners of League of Legends Matches

Abstract: League of Legends is one of the largest online multiplayer games played by millions around the world. This report has studied the accuracy and efficiency of five different machine learning algorithms in predicting the winner in a match of League of Legends with data collected ten minutes after it started. The five algorithms that were chosen for analysis are: Naive Bayes, K- Nearest Neighbors, Decision Trees, Logistic Regression and Support Vector Machines. The analysis of the models showed that they generally scored similarly, except for Decision Trees which achieved a score noticeably lower than the other models. Furthermore when observing the accuracy of the models it could be seen that Logistic Regression performed marginally better than all of the others. However, k- Nearest Neighbours was the model that had the fastest fit time for the biggest training sizes measured while Support Vector Machines were the slowest. The models all achieved an accuracy above 69%, giving the impression that there is a use case for these models in prediction of League of Legends matches.