betsfiled.THE NUMBERS BEHIND THE GAMETomorrow’s picks ↗
THE BETSFILED MODEL

Dixon-Coles football
prediction model.

A transparent football score model trained on real completed matches, with recent form weighted more heavily and a correction for low-scoring outcomes.

How the Dixon-Coles football prediction model works

Betsfiled trains a separate model for each competition using completed league matches. Provider-specific team names are normalised to stable canonical keys, so free historical API-Football results can be matched to current football-data.org fixtures. Every club receives an attacking strength and a defensive strength. A home-advantage term and the competition scoring level turn those strengths into expected home and away goals.

Recent matches matter more

Results are exponentially weighted with a default time-decay value of 0.0019. A match roughly one year old therefore carries about half the influence of a new match. Older results remain useful, but they cannot dominate a team that has changed.

Why Dixon-Coles is different from standard Poisson

Football contains more low-scoring dependence than two fully independent Poisson variables describe. The fitted rho correction adjusts 0–0, 0–1, 1–0 and 1–1 before the score matrix is normalised. From that one coherent matrix we calculate correct score, home-draw-away, BTTS, over/under and double-chance estimates.

Regularisation and promoted teams

Ridge regularisation pulls teams with limited evidence toward the league average. A club without enough usable history is not assigned an invented strong rating; the system falls back to the simpler venue model or withholds the forecast.

Market information

When genuine pre-match 1X2 odds are captured, their margin can be removed and the resulting probabilities compared with the model. A blend is allowed only from timestamped prices. Betsfiled does not reconstruct or fabricate historical odds.

Testing without looking into the future

A proper walk-forward test trains only on matches available before each prediction window. Log loss, Brier score and calibration are evaluated on later matches. Hyperparameters are selected on a validation period and the final report must use a separate holdout period. Synthetic tests verify the code; they are not advertised as real forecasting performance.

What the model cannot promise

No football prediction algorithm guarantees a correct score or profit. Line-ups, injuries, tactical changes, weather and market movement can add information not present in the historical results. Every forecast remains an estimate, and losing selections stay in the public results archive.

Correct score predictions ↗Published results ↗Full methodology ↗
A LITTLE MORE CLARITY

Good questions.
Clear answers.

Understand the predictions before you explore the picks.

Our full methodology ↗
What is the Dixon-Coles model?

It is a football score model that estimates each team’s attacking and defensive strength and adjusts the relationship between low-scoring results such as 0–0, 1–0 and 1–1.

Why are recent matches weighted more heavily?

Team strength changes. A time-decay weight lets newer results influence the fitted strengths more than older matches without discarding useful history.

Does Betsfiled use bookmaker odds?

Only genuine pre-match prices captured with a timestamp can be compared or blended. Historical odds are never invented, and model-only probabilities remain clearly labelled.

How often is the model trained?

Training is refreshed when new completed matches enter a league dataset. Every published forecast stores its model version, training cutoff and original inputs.

Does backtesting guarantee profit?

No. Walk-forward tests measure forecasting quality on unseen matches. They do not guarantee future accuracy or betting profit.

Are Betsfiled football predictions free?

Yes. Published match previews, market selections and results are free to read. No registration or paid prediction subscription is required.

How are the predictions calculated?

Where enough mapped history exists, a Dixon-Coles model learns separate attack and defence strengths with extra weight on recent matches. Otherwise a clearly labelled Poisson fallback uses current home and away records.

Do predictions guarantee a winning bet?

No. A forecast describes possible outcomes, not certainty. A model probability is not a verified win rate, and losing selections remain visible in the results archive.

When are match predictions updated?

Eligible forecasts are generated before kick-off when enough data is available. The original published selection is retained. Scores are refreshed on a schedule, so this is not a real-time livescore service.

Where do the football statistics come from?

Current fixtures and tables can come from football-data.org, while free API-Football seasons from 2022–2024 can supply model history. Keys remain separate, coverage depends on each account, and Betsfiled generates the model output and commentary.

Which timezone do the fixtures use?

Fixture dates and page filters use UTC for a consistent archive. You can choose local kick-off times using the timezone selector; the selected page date remains labelled in UTC.

Where can I check previous predictions?

The Results page lists the original prediction, final 90-minute score and outcome. Extra time and penalty shootouts are excluded from normal-time market settlement.