Prediction of eSports match outcomes using regression models

This bachelor’s thesis focuses on the analysis and prediction of match results in the eSport game League of Legends (LoL) using machine learning. We compare predictive methods used in traditional sports and in eSport to highlight the key differences and specificities of these two fields. We utilize three machine learning methods, gradient boosting, random forest, and logistic regression, to create an accurate and robust model for predicting match results in LoL. Our analyses showed that gradient boosting is the most accurate and robust method among these approaches. Data for our models was obtained from an eSport database and was processed and analyzed in Python using libraries for machine learning and data analysis. During the analysis of the dataset, we encountered several issues, such as the temporality of data and the selection of variables. Nevertheless, we identified advantages of eSport prediction compared to traditional sports, such as the automatic and accurate generation of factors during gameplay. The results of our models were evaluated using metrics such as accuracy, precision, recall, and F1-score. Gradient boosting achieved the best results with the highest accuracy, precision, recall, and F1-score. Overall, our work provides valuable insights into the field of eSport prediction and its comparison with traditional sports. In the future, it would be interesting to examine additional factors that influence match results, such as the preferences of individual players and teams or the impact of game updates on predictions. Our results also emphasize the importance of combining different types of data and their proper balance to achieve optimal results in predicting match outcomes in the eSport environment.