Enhancements of Pass Play Quantification Method with Geometric Features of Formations

In the last two decades, the practical use of player position data has attracted a great deal of attention for analyzing and visualizing plays and tactics in many team sports including soccer, American football, basketball, and rugby. With this background, we proposed a machine learning-based method to quantify pass plays via multiple geometric feature values of formations computed from player position data. This method was successfully applied against a real soccer dataset, but there has still been room for enhancement. In this study, we propose two enhancements of the pass play quantification method to overcome its limitations. The first enhancement is an improvement in modeling by adding pass play feature values as inputs. The improved models are also validated by using a larger real soccer dataset than in our previous work. The second enhancement is related to the visualization for quantification results of the effectiveness and risk for potential pass plays from the detected ball possessor. This visualization is also applied to a real soccer dataset using our visualization tool.