Skip to main content
Department of Information Technology

Statistical methods and machine learning in football

Several models have been proposed to model the quality of play in football. One of these is expected goals which uses logistic regression to evaluate the probability of a shot from a particular location being a goal. Another technique evaluates passes, again using non-linear regression to predict the probability of a particular pass leading to a goal. We currently use this model in the application Twelve (twelve.football) and in analysis at Hammarby football club.

All of these approaches build on data of what happens to the player with the ball, i.e. the co-ordinates of passes and shots. They don’t account for the positions of the other 21 players. We now have tracking data of all of the positions of players during the match in Allsvenskan. The aim of this project is to improve the accuracy of these models using that data. We will try both logistic regression and neural network approaches to see how tracking data can be used to improve the accuracy of these models.

We will also look at ways to visualise the results.

The project will be in collaboration with the company Twelve and Hammarby football club, under the supervision of David Sumpter.

Updated  2019-09-19 17:56:17 by Maya Neytcheva.