Predicting successful player alignments in volleyball using machine learning

This bachelor thesis investigates the application of machine learning techniques to analyze and predict the effectiveness of defensive formations in volleyball. Using a manually annotated dataset of 300 plays from professional volleyball matches, three different machine learning models-logistic regression, random forest, and XGBoost-were developed and evaluated. Among these, the XGBoost model achieved the highest accuracy, but all models demonstrated solid performance, suggesting that even with a relatively small and structured dataset, meaningful predictions can be made. The study emphasizes the importance of feature selection and data preparation, focusing on variables available before the attack, such as defender positions, attack type, set quality, and block details. Feature importance analysis revealed that the type of attack, block effectiveness, and especially the position of defender 6 are crucial factors influencing defensive success. The results show that optimal defensive positioning is not static but should adapt dynamically depending on the specific game situation, such as whether the block touches the ball. Although the dataset was limited in size and scope, this research demonstrates that machine learning can provide valuable tactical insights for coaches and players. With larger datasets and automated data collection methods, these approaches have the potential to support real-time decision-making and strategy development in volleyball, ultimately contributing to improved team performance and a deeper understanding of the game.