📊 How Our AI Predicts Matches

The Core Model

We use a Dixon-Coles Bivariate Poisson model anchored by Elo ratings — a proven statistical approach used in academic football research since 1997. Unlike simple win/loss/draw percentages, this model generates a full probability matrix for every possible scoreline, then derives match outcome probabilities from those score probabilities.

Why Dixon-Coles? Standard Poisson models underestimate low-scoring outcomes (0-0, 1-0, 0-1, 1-1) — exactly the results that matter most in tournament football. The Dixon-Coles correction adjusts for this, producing more accurate probabilities for draws and 1-goal margins.

The 6-Step Prediction Pipeline

1. FIFA Ranking → Elo Rating — Each team's FIFA ranking is converted to an Elo score. #1 = 2400, ±8 Elo points per rank. This establishes the mathematical baseline for team strength comparison.
2. News Modifier — Latest injury reports and lineup news adjust each team's Elo rating by ±5–15 points per impact score. Missing a star striker? Your attack rating drops. Key defender out? Your defensive Elo takes a hit.
3. Goal Trend Analysis — Recent warm-up and competitive match performances are analyzed for goal differential trends. Recent results are weighted more heavily using exponential decay — a match 7 days ago matters much more than a match 3 months ago.
4. Situational Factors — We apply four adjustments: Home Advantage (+40 Elo for host nation USA), Host Nation Bonus (+20 additional Elo for all host nations — USA, Canada, Mexico), Cross-Continent Travel Penalty (negative adjustment for teams traveling across 5+ time zones), and Knockout Pressure Adjustment (teams with more knockout experience get a small boost in elimination games).
5. Dixon-Coles Poisson Matrix — Expected goals (xG) for each team are fed into a bivariate Poisson distribution. This produces a full score probability matrix (0-0, 1-0, 0-1, 1-1, 2-0, ...). The Dixon-Coles correction term (ρ, rho) adjusts for the empirical under-prediction of low scores.
6. Champion Prediction — Adds squad market value (Transfermarkt data), group opponent strength, and host nation bonus to rank all 48 teams by tournament victory probability. The simulation runs 10,000 tournament brackets to produce stable probabilities.

Key Parameters

Data Sources & Update Frequency

Editorial Independence

Our model is strictly mathematical. We do NOT accept money from bookmakers, betting sites, or sports organizations to influence predictions or rankings. No sponsor can pay for a higher prediction.

⚠️ For entertainment purposes only. This site does NOT provide betting advice, gambling tips, or financial recommendations. All predictions are free, public, and should be treated as entertainment.

Limitations

• Football is unpredictable — No model can perfectly predict football outcomes. Our model provides probabilities, not certainties.
• Elo rating lag — FIFA rankings update monthly at most. A team that has dramatically improved or declined between rankings may be misrepresented.
• Injury news quality — News-based modifiers depend on the accuracy and timeliness of public reporting.
• Tournament-specific factors — Penalty shootouts, red cards, and other discrete events introduce randomness our model cannot fully capture.
• Not betting advice — Our predictions are for entertainment and informational purposes. Never use them as the basis for gambling decisions.