Využití statistických metod pro predikci sportovních výsledků

The thesis is focused on the creation of models for predicting the results of NBA basketball games. With the help of historical statistics from the matches and team evaluation based on the Elo model, the created models determine the probability of the home team winning games. There are two ways which have been chosen to create team evaluation based on the Elo model. One of them was the adoption of an already existing modified version of the original model. The second was the creation of the own modification of the model with the help of simulations, where the selection of the optimal values of the model parameters was done by comparing the mean squared errors. The following goal was to find a model that could most accurately determine the probability of the home team winning. The models differ either by using a different modification of the Elo model or by choosing different classification methods. The classification methods chosen for this work are logistic regression and the support vector machines. The model using the adopted modification of the Elo model and logistic regression was chosen as the most adequate. Moreover, the accuracy of its predictions is almost equal to already existing models.