Feature Analysis to League of Legends Victory Prediction on the Picks and Bans Phase

Victory prediction in online video games has become an important application for machine learning due to the large amount of data generated by these games and their growing popularity. The creation of professional leagues also drives these applications, as teams want to know their chances of victory and know what are the determining factors to achieve it. Thus, in this research, we analyze whether pre-game information can be explored for victory prediction of professional matches of League of Legends (LoL), one of the main MOBA games. In experiments, we benchmarked different feature sets and algorithms to assess the victory predictions in LoL. The results show that historical performance information is the most accurate features for performing this task. The induced models, especially Random Forest and Logistic Regression, achieved AUC values of 0.97.