Goalii Methodology
Goalii publishes the methodology behind every prediction so the model can be audited, not just trusted. This page summarises the data Goalii uses, the model family it sits in, and how Goalii measures performance. Detailed backtest tables and the live dashboard are accessible inside the app.
Data sources
Goalii uses licensed match and event data covering top-tier football competitions in Europe, the Americas, and the major international cups. Data is refreshed in near real-time during match windows and on a daily cadence between matchdays.
Full provider list and refresh cadence: to be published.
Model architecture
Goalii uses an ensemble model: a gradient-boosted base learner for goal-market predictions (Over/Under, BTTS) combined with a dedicated probabilistic component for the 1X2 outcome. Live signals (lineups, weather, injuries) reweight the base prediction in the run-up to kickoff.
Feature list, hyperparameter ranges, and ablation results: to be published.
Backtest framework
Every Goalii prediction is logged at issue time with its market, confidence, and timestamp. Once the match concludes, the result is matched back automatically. Hit-rate is computed per market and per confidence band on a rolling 90-day window. Goalii does not retroactively edit predictions — the history shown in-app is exactly what was published.
Backtest tables and prediction-level audit log: to be published.
Limitations Goalii is honest about
- Early-season fixtures have less form data; confidence is reduced accordingly.
- Cup competitions with rotated lineups are harder to model than league fixtures.
- Goalii does not predict in-play (live) markets; predictions are pre-match.
- Past performance is informational and does not guarantee future results.
Where to go next
Last updated