Method
How our predictions work
We turn recent match results into probabilities for a home win, a draw and an away win. They are estimates, not tips, and we publish the misses next to the hits.
The short version
For every fixture, the model looks at each team's last few results: goals scored and goals conceded, with the most recent games counting most. It blends that with the average scoring in the competition and works out how many goals each side is expected to score. A Poisson model then gives the probability of every scoreline, and adding the scorelines up gives the home win, draw and away win probabilities.
What the numbers mean
If a match shows 50% home win, 28% draw and 22% away win, the model expects that kind of match to end in a home win about half the time over many similar fixtures. It does not mean the home side will win. A 22% outcome happens roughly one match in five.
| What you see | What it is |
|---|---|
| 1 / X / 2 | Home win, draw, away win probabilities. They add up to 100%. |
| Forecast score | The single most likely scoreline. Even the likeliest scoreline usually has a low chance, often under 15%. |
| Over 1.5 / 2.5 goals | Chance the match has more than that many total goals. |
| Both teams score | Chance both sides score at least once. |
| Expected goals | The model's average goal estimate for each team. |
What data we use
Completed results from our data provider, football-data.org: roughly the last 60 days, plus a longer saved archive when available. For each team we use up to its last 8 results, and each older game counts about 16% less than the one after it. Where a team has at least 3 home (or away) games, its home (or away) form is used; otherwise its overall form. If the two teams have met at least 3 times recently, the latest meetings nudge the estimate slightly (10% weight). The detail line on each match tells you what the forecast rests on: team-specific form, partial history, or competition averages.
When we hold a forecast back
If there is not enough data for the competition, we show the fixture without a forecast. We would rather show nothing than a number built from a fixed default. Where a team has fewer than 6 recent results, the forecast is marked as partial and pulled towards the competition average.
How we check ourselves
Forecasts are saved before kickoff and graded against the final score afterwards. The results page lists every graded match, including the wrong ones, with the share of correct picks and a confidence range. Small samples are noisy, so treat early figures with caution.
Where the model falls short
- It does not know about injuries, suspensions, lineups, travel, weather or motivation.
- Draws are hard to call. The model rarely makes a draw the most likely outcome.
- Teams with few recent matches, new signings or a new coach are less reliable.
- Odds from bookmakers include information we do not have. This site does not compare against them.
Predictions are for information and entertainment. They are not betting advice and no outcome is guaranteed. You must be 18 or older to bet where it is legal, and you should never stake money you cannot afford to lose.