Abstract
In professional football a key factor regarding injury is the time to return to play. Accurate prediction of this would aid planning by the club in the event of injury. It would also aid the club medical staff. Gaussian processes may be used for machine learning tasks such as regression and classification. This study determines whether machine-learning methods may be used for predicting how many days a player is unavailable to play.
A database of injuries at one English Premier League Professional Football Club was reviewed for a number of factors for each injury. Twenty-five variables were recorded for each injury, including time to return to play. This was determined to be the response variable. We used a Gaussian process model with a Laplacian kernel to determine whether the return to play could be predicted from the other variables.
The root mean square error was 13.186 days (S.D.: 8.073), the mean absolute error was 8.192 days (S.D.:13.106) and the mean relative error 171.97% (S.D.:75.56%). A linear trend was observed and the model demonstrated high accuracy with greater errors being observed for cases where the value of the response variable was higher, i.e. in those cases where the time to return to play was lengthy.
This is the first step in attempting to design a computer-based model that will accurately predict the time for a professional footballer to return to play. The model is extremely accurate for most cases, with errors increasing as the severity of the case increases too.
