Enhancing decision-making in football players using predictive analytics

Abstract

The purpose of the thesis was to enhance football players’ decision-making using predictive analytics and computer vision. The research aimed to support Calevala Interactive Oy by developing a system that analyses player behavior from match videos to improve tactical awareness and reduce in-game decision-making errors. The study focused on three research questions: how predictive analytics improve real-time decision-making, which performance metrics are most influential, and how effective AI-driven training is in practice. The thesis was commissioned by Calevala Interactive Oy. This thesis takes a practical approach. It begins by reviewing important concepts such as machine learning, computer vision, and sports analytics. The project used object detection models to collect data from professional football match videos. Computer vision techniques helped extract player and ball movements. The data was then grouped using unsupervised learning methods, specifically clustering. K-Means clustering and the Elbow Method were used to find player behavior patterns. These results were visualized with heatmaps and movement tracks to support tactical training. The research demonstrates that video-based predictive analytics can effectively improve decision-making by recognizing patterns in player and ball movements. The analysis indicates that metrics like positioning, possession time, and player proximity are key to successful decisions. Based on the findings, it is recommended that football organizations integrate video-based AI tools into their training to support more informed, data-driven strategies. Feedback from the commissioner confirmed that the system offers valuable support for ongoing development and real-world application.