Soccer Dataset: Large-scale Generation of Gameplay Data using the RoboCup Soccer Simulator

Abstract

In team sports with a high degree of freedom of movement such as soccer, the cost of collecting data on play content is high, and it is difficult to prepare sufficient data for analysis. This study addresses this issue by generating large amounts of soccer play data through computer simulation. The system developed in this study uses the RoboCup soccer simulator to generate match data that reproduces the complexity of soccer competitions. Furthermore, soccer game event data is automatically extracted from the object tracking data recorded by the simulator. The data set comprises 78,000 matches, which is equivalent to approximately 8,667 human soccer matches.