JOURNAL OF ARTIFICIAL INTELLIGENCE RESEARCH - 2021
Karl Tuyls, Shayegan Omidshafiei, Paul Muller, Zhe Wang, Jerome Connor, Daniel Hennes, ian Graham, William Spearman, Tim Waskett, Dafydd Steele, Pauline Luc, Adria Recasens, Alexandre Galashov, Gregory Thornton, Ro-muald Elie, Pablo Sprechmann, Pol Moreno, Kris Cao, Marta Garnelo, Pra-neet Dutta, Michal Valko, Nicolas Heess, Alex Bridgland, Julien P ́erolat, BartDe Vylder, S. M. Ali Eslami, Mark Rowland, Andrew Jaegle, Remi Munos,Trevor Back, Razia Ahamed, Simon Bouton, Nathalie Beauguerlange, Jack-son Broshear, Thore Graepel, Demis Hassabis
https://discovery.ucl.ac.uk/id/eprint/10129726/7/Graepel_12505-Article (PDF)-26921-1-10-20210506.pdf
Statistical learning is the foundation of machine learning. It uses statistics and algorithms to help find relationships and patterns within data. In football this would help with making predictions on more complex data.
Due to having low goals (compared to other sports) and not many salient events and many players with multiple roles, football can be quite complex and hard to analyse. Statistical football analysis focuses on understanding and comparing the playing styles of players and teams such as identifying unique traits of how a player/team performs, measuring impact on the pitch and predicting future action and imagining counterfactual scenarios.
Traditionally this was done by aggregating statistics/ on ball actions such as goals, tackles, shots on target etc but now advanced methods attempt to automatically learn the deeper patterns from data. This is because actions off the ball such as team positioning/ movement and whats actually happening on the field are just as important to the game and much harder to capture in current modals. The goal of introducing statistical learning is to learn from layers actions which are not seen until much later in the game and finding out which data is the most meaningful and impactful for clubs. Using reinforcement learning and deep learning will help to create a much better model when it comes to the on pitch flow and strategy in football. This can already been seen in sports such as ice hockey (Liu, G., & Schulte, O. (2018). Deep reinforcement learning in ice hockey for context-awareplayer evaluation. In (p. 3442–3448)
Game theory is the study of strategic decision making between individuals or teams. In football it is used to help analyse and predict players' choices such as whether they shoot left or right in a penalty or set pieces as these are easy to model and study.
Penalties are a common focus when using game theory as it only require 2 players, a clear strategy of left, right or centre and the conditions of them are consistent across the game making it easy to analyse. However live play during a match is much more complex due to 22 players being involved, multiple strategies and each players actions/ decisions can be unpredictable. Therefore more advanced approaches will be needed.
Behavioural game theory helps analyse the more realistic situations such as when players are influenced by psychology, pressure or other external factors and dont perform rationally.