Markov Decision Process modelling of football attacks and analysis of the optimal actions

Football is the most popular sport in the world and continues its expansion to other continents than South America and Europe. The football teams are therefore always looking for ways to improve. They try to recruit the best players and coaches to achieve their objectives. Moreover, with the increasing use of data in the world and in the world of sports they also use it for different purposes. One of them is the performance analysis and optimisation which is the topic of this thesis. This master thesis aims to obtain the optimal actions to increase the probability of scoring in different situations. In order to achieve this objective, football attacks will be modelled as a Markov Decision Process. This model, simple at the beginning, will be improved by taking into account different elements of the possessions, the players' strengths or their location. The transition matrices will be computed using datasets with data collected from real matches. The optimal actions found by the model will be compared to real events to correct or approve the players' choices. Finally, a comparison between the players will be done to find out those performing the actions the closest to the optimal actions.