Potential Pass Networks

Sammanfattning: Passes are the events that happen the most often in football. Therefore, it is important for scouting and situation analysis to recognize the most valuable passes in crucial situations. Aggregating the most optimal passes during the game can provide insights to the coaching staff regarding players' availability to receive dangerous passes, missed dangerous passing opportunities, and pass combinations that should have occurred more or less frequently. In the study, we introduced potential pass networks, which visualize the most valuable passes in the form of previously used pass networks. We implemented these networks with ORTEC event data and TRACAB tracking data using a physics-based approach, incorporating a closest defender model and the concept of error associated with the pass, and convolutional neural networks. Additionally, we compared them between each other.  The physics-based approach yielded a more accurate approximation of the most valuable pass in various scenarios, while maintaining the explainability of the outcome. Machine learning method tended to overestimate the value of long passes close to the sideline and failed to provide reasons behind the bias. Overall, using the physics-based approach, we produced a useful tool which can be used in football clubs.