The AI ​​model tells you why Brazil is most likely to win the championship this year! Accurately predicted the previous two championships

Speaking of the biggest Internet celebrity of the 2010 World Cup in South Africa, it must be "Octopus Paul"!

This magical octopus located at the German Marine Life Center not only successfully predicted the results of all seven games of the German team, but also successfully selected the final champion, the Spanish team.

Unfortunately, Paul has left us forever, but his "legacy" lives on in people's attempts to predict the outcome of football matches.

At The Alan Turing Institute, as the 2022 World Cup in Qatar continues, three researchers, Nick Barlow, Jack Roberts and Ryan Chan, decided to use an AI algorithm to predict this year’s World Cup. The championship belongs.

Predictive Model

The statistical algorithm of Turing Institute is based on a model "AIrsenal" derived from the football game "Premier League Fantasy" (not Wrong, as you can tell from the name, the inventor of this algorithm must be an Arsenal fan).

This model draws on the classic football prediction model launched by Dixon and Coles in 1997. The model takes into account a team's attacking strength, defensive strength and home advantage, and uses Bayesian statistics to calculate which team is most likely to score in a match.

Three researchers adjusted the AIrsenal model to make it more suitable for predicting the results of international competitions.

For example, each national team most often plays against teams from the same continent, such as Brazil, which has not played against any European team since 2019. This can create bias when predicting results between teams from different continents.

In order to solve this problem, the researchers introduced the "relative advantage parameters" between federations on different continents and revised the model.

Training data

In order to accurately predict the winner of the 2022 World Cup, you first need to use past data to train the model.

Three Turing Institute researchers used a comprehensive database of international football matches compiled by GitHub user martj42 - which records detailed results of every football match since 1872.

In the training data, the weight of the World Cup main matches (after 2002) and recent matches are sequentially greater than those of the intercontinental championships, qualifiers and friendly matches. The researchers also input FIFA official rankings into the model to provide an up-to-date estimate of team performance.

It is worth mentioning that the researchers also stated in the press release that their model did not take into account factors such as "players", "penalty kicks" and "location/weather/others", but still Successfully predicted the winners of the 2014 and 2018 World Cups, Germany and France.

Prediction results

In the end, through this AI model and 100,000 experimental results, the Alan Turing Institute predicted the "ultimate championship" and "the results of the England and Wales team" respectively. , and finally two reports were given.

Final Champion Prediction

#We can find from the above table that among the 100,000 test results, Brazil has nearly 25,000 Winning the final championship, the probability of winning is about 25%.

Belgium and Argentina are the second and third favorites to win the championship respectively, with probabilities of 18% and 13.5% respectively.

In addition, the teams ranked 4-10 in probability of winning are France (11%), England (7.5%), Spain (4.5%), Netherlands (4.4%), Denmark (3.1%), and Portugal (3.0%) and Croatia (2.2%) - all European teams.

Performance prediction of the two brothers of Great Britain

According to the AI ​​model simulation of the results of the England and Wales teams, Wales has more than half the probability (52%) Stopped in the group stage.

England is in a slightly better situation, with only a 20% probability of missing the group stage. However, the model still believes that they have a 70% probability of stopping in the top 8, and is not optimistic about the "Three Lions"'s prospects of winning the championship.

I wonder if English fans as far away as the British Peninsula would agree with such a prediction.

Other models are also optimistic about Brazil

Coincidentally.

Recently, a joint research team from the University of Luxembourg, the Technical University of Munich and the Technical University of Dortmund also used data from the five World Cups in 2002, 2006, 2010, 2014 and 2018, combined with parameters such as the structure of each team and the GDP of each country. The "Random Forest Model" was trained to predict the winner of the 2022 World Cup.

The final results show that the Brazilian team has a 23.5% probability of entering the finals and a 15% probability of winning the championship, ranking first among the top 32. The teams ranked 2-5 with the probability of winning are: Argentina, Netherlands, Germany and France.

The joint research team’s AI model’s prediction of the final winner is exactly the same as the research results of the Alan Turing Institute. From this point of view, the crowning of the sixth star of the "Samba Army" is really just around the corner!

Finally, we must also admit that although it is scientific to use AI to predict the results of the game, and teams such as Brazil, Belgium and Argentina are indeed the biggest favorites in this competition, it is absolutely impossible for the model to combine all factors All involved.

No matter how accurate the AI ​​model is, football will always be a random game.

The unpredictable factors on the green field may be the reason why we like it and love it deeply.

Reference:

https://www.zeileis.org/news/fifa2022/

https://www. eurekalert.org/news-releases/971175

https://www.turing.ac.uk/blog/can-our-algorithm-predict-winner-2022-football-world-cup

https://www.iflscience.com/ai-has-run-100-000-simulations-and-predicted-the-2022-world-cup-winner-66314

The above is the detailed content of The AI ​​model tells you why Brazil is most likely to win the championship this year! Accurately predicted the previous two championships. For more information, please follow other related articles on the PHP Chinese website!