Football Predictions
& Data API
Built for developers
Access football predictions, probabilities, fixtures, live data, odds and advanced match analytics through a production-ready API. Scored by the PredictIQ Stack ensemble (CatBoost, LightGBM, a Poisson goal model and market-implied odds), over REST or straight into your model via MCP.
- Live accuracy · 3-way (home/draw/away)
- 46.0%
- Matches evaluated
- 3,807
- Leagues covered
- 88+
A random guess scores about 33% on a three-way market. Calibration, Brier score and log loss are published on the accuracy page.
Real numbers, measured against results — see the full breakdown
One of 40+ drop-in widgets — browse them all
Not a black box, not a guess
Never gets worse
A retrained model only goes live if it beats the one it's replacing. A scheduled retrain can never silently make predictions worse.
Calibration, not just accuracy
We track whether a stated 60% actually happens 60% of the time — the number that matters if you're building on these probabilities.
No inflated claims
Every market — 1X2, over/under, BTTS — comes from the same scoreline model, so they never quietly contradict each other.
Live Match Radar
Our stacked model re-scores win probability continuously while these matches are in play — the same number, updated as the game moves, not a fixed pre-match figure.
Nothing kicking off right now — check back once a match is live.
The starting XI
Five components, five jobs. No single model makes the call alone — each one reports its read, and the captain decides.
Poisson xG model — Home
Projects the home side's expected goals from recent shot quality — reads where a chance is likely to fall, not just who has been winning.
Poisson xG model — Away
The same read, mirrored for the away side — a separate model trained on its own, since home and away scoring patterns aren't symmetric.
CatBoost
Reads pattern across form, rest days, head-to-head history, venue-specific form, and the market's own price, all at once.
LightGBM
Same inputs as the playmaker, a different tactical read — the disagreement between the two catches what either would miss alone.
Meta-learner
Weighs the other four models' reads alongside devigged market odds and makes the final call. Only replaces the live model if it proves better, never worse.
Why stack five models
No single read decides the call. A retrained model only replaces the one you see here if it beats it — a scheduled retrain can never quietly make production predictions worse.
The command centre
A live look at what the model is watching, right now.
AI Win Probability
previewPremier League · Man City v Aston Villa
Model Performance
Confidence for this match vs. real published accuracy
Model Status
predictiq_stack_v746.0%
live accuracy
3,807
matches evaluated
Expected Goals
model-impliedMan City
1.72Aston Villa
0.88Powerful football data
Live scores
Fixtures and live match updates, updated in seconds.
Leagues & standings
Competition tables and rankings for 80+ leagues.
Teams
Squads, statistics and fixtures for every club.
Players
Profiles, statistics and career data.
Match statistics
Events, lineups and detailed in-match statistics.
Predictions
Probability-based football analytics for every fixture.
Football widgets
Add live football data to your website with a script tag.
Model Breakdown
Secondary Markets
Where the market's best price is right now
We compare live prices across every bookmaker in our feed and only surface a selection once it's genuinely ahead of the field.
Models stacked
5
Leagues covered
88+
Live coverage
24/7
Start free, upgrade later
No contracts. Cancel anytime from your dashboard.