As excitement builds up for the UEFA Euro 2024, data scientists predict highest winning chances for France, using refined machine learning.
As millions of fans await Friday’s kickoff of the UEFA Euro 2024, the big question is: who among the teams will succeed, who will drop out, and who will eventually prevail?
Since 2018, Prof. Christophe Ley from the Department of Mathematics at the University of Luxembourg, together with a group of European scientists, works on machine learning combinations to predict outcomes of all possible matches in football tournaments, and therefore winners of the tournaments. This year, he has been joined in this endeavour by his doctoral candidate Florian Felice from the Department of Mathematics. To obtain consistent results, researchers use a two-step approach.
In a first step, three complex statistical models assess the strengths of teams and individual players, using data such as historical match data, odds from 28 bookmakers and player ratings, drawing on individual performances at both home clubs and national teams.
Then, an AI model under the form of an ensemble of machine learning models is trained on the results of UEFA Euro tournaments from 2004 to 2020 to decide how to best combine the outcomes of these statistical models with other information about the teams such as market value, number of Champions League players and the home country’s GDP as a socio-economic factor.
Once the AI trained, the researchers forecast outcomes for the Euro 2024. By estimating the predicted number of goals for all possible matchups, the model can determine the probabilities of wins, draws, or losses for each game using a distribution called bivariate Poisson. This approach allows for the simulation of the entire group phase, identifying which teams will advance to the knockout stage and ultimately, the winner. The researchers simulate the tournament 100,000 times and calculate for every team the percentage of times that they come out as winners of the tournament.
And this year?
“France comes out as the top contender for the European title, followed by England and Germany”, says Christophe Ley. Indeed, their model gives France a winning probability of 19.2%. England follows closely with 16.7%, and host nation Germany holds a 13.7% probability of securing the cup.
Further applications are foreseen: “Florian is currently setting up a similar AI model for the Olympic Games in both football and handball”, adds Christophe Ley. “Handball match predictions are part of his PhD thesis, and we already wrote a paper about predicting matches of female handball clubs.”
Besides Christophe Ley and Florian Felice, the research group includes Prof Andreas Groll and Jonas Sternemann (Technical University Dortmund), Prof Lars Magnus Hvattum (Molde University College), Dr. Gunther Schauberger (Technical University of Munich) and Prof. Achim Zeileis (University of Innsbruck).
Photo at the top of the page: © EnvatoElements, LightFieldStudios