Making the cut: forecasting non impact injuries in professional soccer
Thesis degree name:
Master in Management + Analytics
This paper proposes a methodology to predict work in non-traumatic injuries in professional soccer players. The task to be solved is a classification problem of the player's status with a window of 72 hours. The data set used corresponds to records of complete training by the players of Belgrano de Córdoba professional soccer team of the first division of Argentina. The chosen model is GBM with an AUC of 0.7. Interpretation exercises based on SHAP are performed on the chosen model to analyze the characteristics that determine the model's predictions. In addition, possible extensions are proposed such as the use of the results of the model at the time of contractual negotiation given the estimated proportion of time that the player will spend outside due to injury and the economic cost of those absences given, at least, by the direct salary cost of that player. Another approach to the injury forecasting problem based on survival time models is also discussed.