Going through a crude mixture of data, econometrics, and machine learning, Goldman Sachs has predicted that Brazil will win the FIFA World Cup in Russia this year.
Goldman Sachs arrived at the answer by firing up machine learning to run 2,00,000 scenarios, mining team data and individual player attributes that enabled them to project-specific match scores. This process has enabled them to project-specific match results after which they have simulated over 1 million variations of the tournament draw to calculate the probable winner.
This isn't the first time their AI has tried to predict a tournament winner. Goldman Sachs had also predicted back in 2014 that Brazil would lift the World Cup. Unfortunately, Brazil lost to Germany in the semi-final, leading to a final clash between Germany and Argentina, which the former won 1-0.
According to the study, France has better chances of winning the cup than Germany but its semi-final match with Brazil is likely to dash all its hopes. Russia although being the host and in a fairly easy group, they are forecasted to crash out of the tournament in the group stages itself.
In the other semi-finals, Germany will finally put an end to Cristiano Ronaldo's campaign to win a World Cup as the Die Mannschaft have better odds against Portugal. Argentina and Spain are expected to underperform, both crashing out in the quarterfinals.
The authors of the study have added a disclaimer as well stating, "But the forecasts remain highly uncertain, even with the fanciest statistical techniques, simply because football is quite an unpredictable game. This is, of course, precisely why the World Cup will be so exciting to watch."
A separate AI run by researchers from the Technical University of Dortmund in Germany predict that Spain has the best chances of winning at 17.8 percent after 100,000 different simulations.
The predictions are based on the standard decision tree model, which suffers from a problem known as overfitting. Overfitting refers to a model that models the training data too well. Overfitting happens when a model learns the detail and noise in the training data to the extent that it negatively impacts the performance of the model on new data.
Source: Goldman Sachs