Written by Gregory Everett. Email: gae1g17@soton.ac.uk
Player tracking data has the potential to drive value for clubs and present new research opportunities in soccer analytics (e.g., physical metrics, space analysis and pitch control). However, this data is extremely expensive due to the advanced data collection process, meaning it is unaffordable to the vast majority of clubs. Therefore, in this work, we present a model to impute snapshots of player tracking data from event-based data which is far cheaper and more widespread. This model consists of a graph network of long short-term memory (LSTM) models and hence captures both the spatial and temporal structure of player positioning. Finally we apply imputed player positions to off-ball analyses such as pitch control and player physical metrics.