Predicting professional League of Legends matches - A comparative analysis of early-game predictors across major leagues

Sammanfattning

This thesis examines whether early-game indicators observed at the 15-minute mark can be used to predict match outcomes in professional League of Legends and whether predictive performance differs across major leagues. Using team-level match data from the LCK, LEC, and LCS, several classification models are estimated to assess the predictive value of early-game performance measures. The results indicate that early-game indicators provide substantial predictive power across all examined leagues. Using only early-game information, match outcomes can be predicted with accuracies in the range of approximately 73–75%. Simpler models perform comparably to, or slightly better than, more complex machine learning approaches, with logistic regression achieving the highest overall performance. Predictive accuracy varies systematically across leagues, with matches from the LCK being more predictable than those from the LEC and LCS. For the LCK, prediction accuracy reaches close to 78%, while corresponding values for the LEC and LCS are lower, typically around 72–74%. This study adds to the existing literature by comparing the predictive performance of early-game indicators across major professional leagues and documenting differences in predictability between regions.