Estimating the most important football player statistics using neural networks

Abstract

Objectively knowing when an individual football player is good can be a difficult task, as football is a team sport and therefore the number of factors that must be taken into account are huge. Human judgements in these situations are therefore often mislead. Studies have shown that neural networks are a useful tool for these situations, due to their ability to extrapolate complex relations between input and output. In this thesis, we ask which player statistics are most important for determining a player’s performance in football. We present two neural networks that attempt to do this by looking at the defending and attacking statistics separately. As a measure of strength of a team we took the numbers of goals the team was expected to score. Results from testing the algorithm showed better results for the attacking model than the defending one. Overall the results looked promising but there is still room for improvement. Future study will definitely need to take into account a player’s competition strength in order to make a better judgement about a player’s objective strength.