Predikční analýza sportovních dat

In the literature, the problem of insufficient datasets in the field of sports data analysis is generally encountered; they are poorly available, short, often outdated, and if actual statistics do appear somewhere, they are not in a form for analysis. Therefore, the first part of this thesis solves the above mentioned problem by building a robot using Python to download statistics from flashscore.com, further transforming these statistics into a dataset suitable for analysis. Furthermore, the work deals with the prediction of match winners based on the statistics just downloaded, namely hockey NHL and ELH, and football FNL, Premier League, Serie A, La Liga and Ligue 1. The variables entering the model for predicting winners consist of historical statistics before the match and information that the bettor sees before the match starts. Using different variations of the XGBoost and Logistic Regression models and then evaluating them and describing the differences between the accuracy of the models across leagues/sports. Subsequently, and thanks to historical information on the odds for a given match, the thesis tries to answer the question whether betting according to the predictions of the established models is a profitable activity in the long run. Or whether there is a particular strategy in betting where betting would be profitable. Then this thesis deals with the evaluation of these strategies based on the identified criteria, thus the output of this part of the thesis is the top 10 strategies that a bettor should follow when betting.