FOOTBALL · MACHINE LEARNING
World Cup outcomes,
modeled and simulated.
A project exploring how team ratings and match predictions translate into tournament probabilities. The model compares past World Cups and simulates the 48-team format used in 2026.
01 / HISTORICAL EVALUATION
How the model performed
The project reports leave-one-World-Cup-out validation and tournament-level Monte Carlo evaluation. XGBoost was selected from the models tested.
| Year | Actual champion | Model's top pick | Reported probability |
|---|---|---|---|
| 2006 | Italy | France | 4.94% |
| 2010 | Spain | Spain | 33.71% |
| 2014 | Germany | Germany | 19.84% |
| 2018 | France | Brazil | 24.38% |
| 2022 | Argentina | Argentina | 20.17% |
Figures are taken from the supplied project description. They have not been independently reproduced for this page.
02 / SIMULATION SNAPSHOT
Reported 2026 probabilities
The project describes one million tournament simulations using a 48-team format, including the best third-place teams and the Round of 32.
01Spain
24.82%
02France
19.67%
03England
14.13%
04Portugal
12.07%
03 / METHOD
From ratings to outcomes
Match predictions become inputs to repeated tournament simulations. The proportion of runs in which a team reaches each stage estimates its probability.
- 01
Prepare team features
EA Sports FIFA game ratings, Elo ratings, and FIFA ranking.
- 02
Predict matches
Compare Logistic Regression, Random Forest, XGBoost, LightGBM, and CatBoost. The project reports XGBoost as the selected model.
- 03
Simulate the tournament
Resolve groups and knockout matches, record the champion, and repeat.
REPRODUCE
Run the 2026 simulation
One million iterations with random seed 42, as documented in the supplied project description.
python -m scripts.run_wc_2026 1000000 42The command and listed outputs come from the project description. Source code, dependencies, and output files were not provided for verification.