The 2018 FIFA World Cup, which took place in Russia from 14th June to 15th July, kept the whole world engrossed at the happenings.
Importantly, soccer enthusiasts and artificial intelligence specialists were looking forward to the team that would be awarded the coveted trophy.
Ultimately, when France played against Croatia in the final match, the former proved to be the best team and were crowned the 2018 FIFA World Cup Champions. Belgium and England took the third position and fourth position respectively.
Earlier AI predictions
Before the World Cup started in Russia, several researchers and scientists tried to predict the outcome of the popular sporting event.
A notable example is a team of researchers from the German Technische Universitat of Dortmund, the Ghent University in Belgium, and the Technical University in Munich who ran an artificial intelligence (AI) algorithm to foretell the likely winners of the tournament.
You can find the results of their findings here.
To predict the possible winner, the AI World Cup prediction researchers simulated the event 100,000 times and used three different data modeling techniques: Poisson regression models, random forests, and ranking methods.
The researchers used data taken from the previous World Cups and analyzed them alongside several other influence variables, including FIFA ranking, team structure, economy of the participating country, and skills of the coach.
Here is a table showing the results of the 2018 FIFA World Cup AI study:
As you can see from the above table, Spain was predicted to be the World Champion with a winning probability of 17.8%, followed by Germany (17.1%), Brazil (12.3%), and France (11.2%).
As we saw in the actual results of the 2018 FIFA World Cup, the AI-simulated results were largely off.
Two of the best predicted teams, Spain and Germany, failed to reach the quarter-finals. Although Brazil reached the quarter-finals, it could not sail to the finals.
Why did the AI World Cup prediction fail?
When carrying out predictions using artificial intelligence techniques, it is essential to have the right data for training and modeling.
Whereas the researchers carrying out the AI World Cup 2018 prediction had the right data and relatively extensive dataset (including history of the past four World Cups) as well as good algorithms with suitable parameters, the AI training models shattered far too early.
This failure could be attributed to the nature of the event that was being predicted.
Just like any other human-based activity, the FIFA World Cup is dependent on several factors that determine the winner of a match after at least 90 minutes of playtime—which the researchers could not sufficiently put into consideration.
To increase the accuracy of the predictions, each single minute of every match could have been simulated.
If the state of each match during every minute were simulated, the reliability of the prediction outcomes could have been increased.
Furthermore, apart from internal factors such as team structure and skills of the coach, other external factors such as unfair referee, players’ personal problems, fans’ sentiments, and weather also play a significant role in determining the outcome of football matches.
The external factors are normally difficult to integrate when performing AI machine learning simulations. Additionally, there are always possibilities of uncertainties happening; for example, when a Croatian player scored an own goal during the final match—which further complicates accurate predictions.
The AI prediction gave a probability of an event and not its certainty.
Although the researchers did not predict France to win the World Cup, they gave it a 11.2% chance of becoming the World Champion—which is not that bad, after all.
In fact, another AI algorithm, called ‘FIFA 18’, correctly predicted France’s win.
Nonetheless, the 2018 FIFA World Cup and other sporting events are some areas that artificial intelligence algorithms need to be improved to enhance the accuracy of their predictions.
Maybe researchers who will be performing simulations in future should integrate wider parameters to strengthen the odds.
Therefore, despite the poor performance of the World Cup 2018 winner models, it shows that AI can be used to make predictions that can escalate processes and increase the efficiency of operations in the world.
If you want to learn how artificial intelligence works to predict outcomes, you can watch hands-on projects taught by professionals in the industry.
For example, AndreyBu, who is from Germany, uses artificial intelligence to accelerate the features of applications. You can learn from one of his practical projects here.