World Cup predictions: How many games did our AI get right?

In total, Kashef, our AI robot, had a 67 percent accuracy level, but failed to predict Argentina as the 2022 World Cup winners.

(Al Jazeera)

World Cup 2022 produced incredible football.

At the start of the tournament, Al Jazeera introduced Kashef, our artificial intelligence (AI) robot, to crunch the numbers and predict the results of each game.

After every day of action, Kashef downloaded the day’s data and compared it with more than 200 metrics, including the number of wins, goals scored and FIFA rankings, from matches played over the past century, totalling more than 100,000 records, to see who was most likely to win the following day.

INTERACTIVE World Cup AI predictor
Al Jazeera developed an AI match predictor game for the 2022 World Cup [Al Jazeera]

Kashef’s pre-tournament predictions

The group stages from November 20 to December 2 were not very kind to Kashef, who erred on the side of caution and failed to foresee any of the many major upsets.

The good news for us sentient beings is that every time Kashef got it wrong, we were treated to a feast of World Cup magic, including Saudi Arabia’s stunning 2-1 victory over Argentina, Morocco’s 2-0 defeat of Belgium and Tunisia’s 1-0 win over 2018 champions France.

At the start of the tournament, Kashef predicted that five-time winners Brazil would beat Argentina in the semifinal and take on France on December 18.

The knockout table below shows who Kashef had reaching the knockout stages at the start of the tournament.

INTERACTIVE Kashef predictor - 21 November
Screenshot of Kashef’s predictions from November 20, 2022 [Al Jazeera]

The knockout table below shows the actual results of the World Cup.

INTERACTIVE Kashef December 18
Screenshot of Kashef’s predictions from December 18, 2022 [Al Jazeera]

Correctly predicted two out of three matches

Like all AI engines, Kashef would generate a probability that a particular team would win, draw or lose a match.

To calculate these probabilities we ran the data through Google Cloud’s Vertex AI, which produces state-of-the-art models – adopting Google Brain – that use Efficient Neural Architecture Search (ENAS) to find the best prediction model. The data pipeline is built with GCP-BigQuery.

Based on our simulations for the 2014 and 2018 World Cup tournaments, Kashef was about 71 percent accurate, with an area under the curve (AUC) of some 67 percent.

To put that in perspective, the random chance for a knockout match is 50 percent – half of the time a team will win and half the time they will lose. During the group stage, correctly predicting an outcome at random is split between three choices – a win, draw and a loss – with a 33 percent chance each.

Predicting match results is not easy. External factors, like team morale or player fitness, make a big difference in how the game goes.

In the quarterfinals, Kashef failed in 50 percent of the games, expecting Brazil to go to the final, while it was also convinced that Portugal would beat Morocco.

In the semifinals, Kashef was 100 percent accurate, while it ultimately failed to predict that Argentina would win the title.

INTERACTIVE Kashef 67 percent
(Al Jazeera)

Who can predict a football game better?

Behind the captivating lure of the footballing spectacle, there was an existential battle taking place at the Al Jazeera offices.

The question: Who can predict a football game better? Kashef, our beloved AI robot, or humans? Throughout the tournament, Al Jazeera conducted a daily poll on our Sports’ Twitter account.

How did Kashef fare in the last four matches?

Argentina vs France (Final)

Kashef’s prediction: France win – incorrect

Al Jazeera’s poll: Argentina win – correct

Final result: Argentina 4-2 (penalties) France

INTERACTIVE_KASHEF_ARGENTINA_V_FRANCE
Kashef’s final prediction with only 1 percent separating Argentina and France [Al Jazeera]

Croatia vs Morocco (Third-place playoff)

Kashef’s prediction: Croatia win – correct

Al Jazeera’s poll: Morocco win – incorrect

Final result: Croatia 2-1 Morocco

INTERACTIVE_KASHEF_CROATIA_V_MOROCCO
(Al Jazeera)

France vs Morocco (Semifinal)

Kashef’s prediction: France win – correct

Al Jazeera’s poll: Morocco win – incorrect

Final result: France 2-0 Morocco

INTERACTIVE_KASHEF_FRANCE_V_MOROCCO
(Al Jazeera)

Argentina vs Croatia (Semifinal)

Kashef’s prediction: Argentina win – correct

Al Jazeera’s poll: Argentina win – correct

Final result: Argentina 3-0 Croatia

INTERACTIVE_KASHEF_ARGENTINA_V_CROATIA_edit
(Al Jazeera)

See Kashef’s predictions during the quarterfinal matches.

How much money did each team receive?

According to FIFA, the total prize money for the 32 participating nations at the 2022 World Cup was $440m.

The breakdown is as follows:

  • Champions: $42m
  • Runners-up: $30m
  • Third place: $27
  • Fourth place: $25m
  • 5th-8th place: $68m ($17m per team)
  • 9th-16th place: $104m ($13m per team)
  • 17th-32nd place: $144m ($9m per team)

In addition, each team would have received $1.5m before the competition to cover preparation costs.

World-Cup-Final-INTERACTIVE-Prize-money-football
(Al Jazeera)
Source: Al Jazeera