Times advance and technologies even more. We have gone from the octopus Paul, who predicted the results of the soccer world cups by selectively devouring the mussels that his keepers placed in a strategic position, next to the flags of the different participating countries, to artificial intelligence. But the latter has yet to prove its effectiveness, applied to a sport in which the unforeseen and the magic of the ball make up a good part of the result. Paul the octopus correctly predicted the outcome of the euro 2008 and, shortly before passing away, World Cup 2010, where he matched the eight team pairings that were proposed to him. The brand is not easy to beat. Those who have accepted the challenge have been a team of specialists from the universities of Innsbruck (Austria), Ghent (Belgium), Luxembourg, and the technical universities of Dortmund and Munich (Germany). Their computers have simulated the Qatar World Cup one hundred thousand times, match by match, following the draw for the tournament and all the FIFA rules, until reaching the verdict: it will be the national team of Brazil that wins the competition.
Brazil, with a 15% chance of winning, would be the favorite team. The Argentina team also has “a lot of chances to win the title” (11.2%). They are followed by the Netherlands (9.7%); Germany (9.2%) and France (9.1%). Only later does it appear the selection of Spain with a 7.5% chance. The forecast combines several statistical models on the “playing strength” of the teams and has been fed with information on the structure of the teams and their players, which includes data such as the market value or the number of players in the Champions League. , as well as socioeconomic factors of the country of origin that have been considered relevant, from the population and its age structure to its gross domestic product.
In each simulation, the system has been learning all the probabilities with which the teams will advance or not to the different rounds of the tournament and the probabilities that they will finally win the Cup or not. The authors warn that they are providing “probabilities, not certainties ». “A 15% chance of winning also implies an 85% chance of not winning,” emphasizes Achim Zeileis, from the department of Statistics from the University of Innsbruck, which avoids getting wet more than necessary. Zeileis points out that similar studies had already been carried out in previous competitions, but the Qatar World Cup It is especially interesting from a scientific point of view due to the unusual date on which it is held: the fact that it is held in winter but with high summer temperatures.
In the winter months, all the major soccer leagues in Europe and South America have had to interrupt their regular match schedule to accommodate the tournament, giving national teams less time to prepare and players less time to recover before and after the event. Extreme weather conditions increase the risk of injury, a variant that has been included in the study. Therefore, “having a team with many players in international leagues, such as the Champions League, la Europa League o la Europa Conference League, it may be more of a disadvantage than an advantage this year,” explained Zeileis, who looks at football from a scientific point of view and is enthusiastic about the mathematical challenge. “All these factors make it more difficult this time to predict how the tournament will turn out, since the variables that were very significant in previous World Cups might not work well or work differently,” adds Andreas Groll, another of the scientists who has participated. in research published by Reports and Proceedings.
The scientific article documenting the study explains that the calculation was based on four sources of information: a statistical model of the “playing strength” of each team, based on all international matches in the last eight years, and another of “playing strength” of game» of the teams, based on betting odds from 28 bookmakers. The third base is additional information about the equipment and the fourth is a machine learning model that combines the different sources and optimizes them step by step.
