In recent years, the analysis of sports matches has flourished, leading to the development of additional statistics on teams and team individuals. This also applies to basketball, a sport that is particularly interesting because of its dynamic character. Spatio-temporal data containing data on ball and player positions at a given point in time have thus been added to the play-by-play data containing data on the events during an attack. Moreover, the number of passes between players, the sequence of passes, and the impact of a pass on the success of an attack represent interesting statistical elements obtained from the data. The master’s thesis provides an analysis examining the changes in pass dynamics and team performance in the absence of the best player. We use two of machine learning approaches: hierarchical clustering, based on the graph of passes between player positions and the graph of passes between court areas, and neural networks, where the success of an attack is predicted based on spatio-temporal data and data obtained from the analyses of pass graphs. The results show that team performance is better when the best player is on the court. What is more important, the results of hierarchical clustering indicate that passes between court areas better predict team performance throughout the season than passes between player positions. The accuracy of supervised learning in predicting the success of an attack was 68,25 %.
|
