BetAttack Prediction Model V1.2
A transparent look inside the statistical model behind BetAttack football predictions. See the data V1.2 analyses, how recent matches are weighted, how expected goals are calculated and how those numbers become match probabilities.
Recent form matters. Context matters too.
V1.2 is built around a simple idea: a team's most recent matches should carry the strongest signal, but those matches should not be read in isolation. A strong performance against a strong opponent means something different from the same performance against a weak one. Home and away context, season performance and the competition's normal scoring level also matter.
The model combines those signals to estimate how many goals each team is expected to score. It then evaluates possible scorelines and converts them into probabilities for the football markets shown on BetAttack.
How a prediction is built
Select each stage to see what happens to the data as a fixture moves through the model.
Last 5 matches
V1.2 reads the five most recent matches available before the fixture. The matches are weighted by recency so the newest result contributes more than the oldest. The recency weights are 1.00, 0.90, 0.80, 0.70 and 0.60.
The V1.2 weighting
Four sources build each team's attack and defence values. The Last 5 signal is already adjusted for the strength of the opponents faced before this blend is calculated.
The primary signal. Recent goals scored and conceded, weighted toward the newest matches.
Recent venue specific performance where at least two relevant matches are available.
Season scoring and defensive averages provide longer term stability.
The normal home or away scoring level in the competition.
Attack meets defence
After the four inputs are blended, V1.2 has an attack and defence value for each team. The home attack is paired with the away defence and the away attack with the home defence.
The model uses a geometric mean rather than a simple average. This reduces the chance that one unusually high input dominates the final expected goal value.
Optional xG enhancement
If both teams have usable recent expected goals and expected goals against data, V1.2 creates an xG based matchup estimate. The existing expected goal value retains 85% of the weight and the xG estimate contributes 15%.
Safety bounds
Final expected goal values are restricted to a range of 0.20 to 4.50 per team. These are calculation safety bounds designed to prevent pathological inputs from producing unrealistic probability distributions.
From expected goals to probabilities
Move the example expected goal values below. The probabilities recalculate using the same Poisson score method used by V1.2. These controls are an illustration of the probability stage, not a live fixture prediction.
Model probability is separate from bookmaker odds
V1.2 produces its statistical probabilities before bookmaker prices are considered. The expected goals calculation does not use bookmaker odds.
Once a model probability exists, available odds can be evaluated separately. BetAttack can convert bookmaker prices into implied probabilities, remove the overround where the complete market is available and compare that market view with the model.
How the model has evolved
BetAttack keeps model versions distinct so methodology changes are visible rather than silently replacing the previous approach.
Last 5 Analysis Model
Current release
V1.2 makes individual recent matches the primary signal. It analyses each team's five most recent matches, weights the newest games more heavily, adjusts for the strength of the opponents faced and combines that information with venue, season and competition data.
Data coverage and confidence
V1.2 analyses a maximum of five recent matches for each team. Its internal confidence label describes data coverage, not the probability that a prediction will win.
What the model cannot know
A statistical model is an estimate of uncertainty, not a description of what will definitely happen. Football contains events that cannot be predicted reliably from historical team data alone.
Unexpected team changes
Late tactical changes, unexpected selections and other information not present in the model data can alter a match.
Red cards and unusual events
A dismissal, injury, penalty or individual error can dramatically change a match after kick off.
Small samples
Recently promoted teams, new competitions or limited historical coverage can provide less evidence than established datasets.
Probability is not certainty
Even a correctly estimated high probability outcome will fail some of the time. That is inherent in probabilistic forecasting.
Prediction model FAQ
No. In V1.2 bookmaker odds are not inputs to the expected goals calculation or the statistical probability model. Odds can be compared with the completed model probability afterwards to calculate implied probability, edge and expected value.
See the model applied to upcoming matches
Browse current football predictions, model probabilities and fixture statistics generated from the V1.2 methodology.
BetAttack predictions are statistical estimates and are not guarantees of future results. Model probabilities can be wrong and individual match outcomes are inherently uncertain. Betting involves risk. Please gamble responsibly and only bet what you can afford to lose.