Achim Zeileis of the University of Innsbruck and his team trained a machine‑learning algorithm to estimate the chances for the 2026 FIFA World Cup, reporting their work in The Conversation. The model operates in two stages: first, statistical models are combined with insights from betting markets and transfer‑market experts to assess team and player strengths. The analysis used all international matches from the past eight years.
Player ratings were derived from their contribution to goals at club and national level, while current quality and future potential were reflected by expected market value. Additional inputs included FIFA ranking positions, the number of players from each country appearing in this year’s Champions League semifinals, and even GDP per capita for the country.
In the second stage a machine‑learning algorithm decides how best to combine the strength estimates with other team‑level information and produces probabilistic forecasts for the outcome of every match.
Examples and the opening match
One concrete output: the AI estimated Mexico would score an average of 1.9 goals in the opening match versus South Africa’s 0.7. That translated to a 65% chance of a Mexico win, 14% for South Africa, and 21% for a draw. The prediction proved correct in this case: Mexico beat South Africa 2–0.
Tournament simulation results
The researchers simulated the full tournament, including the draw and match rules (extra time and penalty shootouts), 100,000 times. The headline probabilities for winning the tournament were:
- Spain: 14.5%
- England: 12.4%
- France: 12.4%
- Germany: 11.2%
- Portugal: 8.9%
- Argentina (defending champion): 8.2%
- Brazil: 4.7%
They also report that, under the expanded 48‑team format, the United States has a 78% chance of reaching the last 32.
Caveats and past performance of models
Zeileis notes that while data‑driven models have had successes — another algorithm correctly predicted the 2019 Women’s World Cup winner — they can also miss. For example, earlier algorithmic forecasts for the 2022 men’s and 2023 women’s tournaments did not necessarily identify the eventual champions as the top favourites. Consequently, the team warns that the 2026 simulation, useful as it is for gauging probabilities, remains subject to errors and uncertainty.
Conclusion
Using a two‑stage machine‑learning approach fed with eight years of match data and multiple market and ranking indicators, the simulation ranks Spain as the single most likely winner of the 2026 World Cup at 14.5%, while also highlighting the substantial uncertainty inherent in such forecasts.
