Datově řízená predikce výsledků tenisových zápasů a její srovnání se sázkovými kurzy

This bachelor thesis deals with the possibility of results predictions in the context of professional tennis matches and its comparison with bookmaker odds. The main goal of the thesis is to verify whether, with the use of machine learning, it is possible to create probability estimates of the match results that give informational advantage compared to the bookmaker odds and to assess their practical use within the context of sports betting. The analysis uses historical data about ATP matches supplemented by information about the betting odds. Based on them are created predictors, describing the performance of players, their form, head-to-head statistics and any game data from previous matches. The prediction models are used by a few selected methods of machine learning– namely: Logistic Regression, Random Forest, Support Vector machine method and XGBoost algorithm. The quality of the predictions is evaluated not only by classification metrics, but mainly by simulation of betting strategies based on probability predictions of the models. The results are compared with implicit probabilities based on the betting odds. The thesis shows that the models, in combination with specific betting strategies, can beat out odds of betting agencies, based on the results of ROI and PnL.