Forecast World Cup 2014 in the language of Wolfram

At the World Cup, the group stage ended, the number of participating teams was halved, and pairs of playoff participants became known. In this regard, the developers of the company Wolfram Research have updated the calculation of statistical probability for the remaining matches.

Previously, about 200 MB of data extracted from specialized sites were uploaded to Wolfram | Alpha. This is, first of all, historical data on 30,000 international matches from 1950 to 2014, including team members and player characteristics. This is not only for predicting the outcome of matches, but for Wolfram | Alpha to process search queries related to football.

For example, here are the statistics on the change in height and body mass index of football players over the past 50 years.
')




Returning to the analysis of the World Cup 2014, the model calculates the probability of the outcome of each match, based on the statistics of past games, including taking into account the advantages of the home field, the advantages of the continent and other factors that emerged after analyzing the historical base. The model then went through training at the last 2000 matches and showed an accuracy of 48%.



After the teams were added to the Elo-rating model, the accuracy increased by 58.3%, and for the knockout matches - 75.7%. This is quite enough to calculate the probability of the outcome of the 2014 World Cup.

So, after the group stage, the probability of winning the tournament is distributed as follows.



Compared with the previous forecast, the chances of the Netherlands have increased dramatically (thanks to a major victory over the favorite).

The most likely pair of finalists is Brazil and the Netherlands.



The entire grid of expected playoff game outcomes is shown in the diagram.



Another set of charts shows the probability of winning (blue) or losing (orange) for the 9 most-rated teams at each stage of the playoffs.



In the group stage of the 2014 World Cup, Wolfram | Alpha forecast accuracy was 62.5% (the engine correctly calculated Costa Rica’s victory over Italy, a draw in a match between Germany and Ghana, a draw by Brazil and Mexico, and several other unobvious outcomes). Let's see how the model will show itself in play-off matches.