How the Model Works

A two-layer statistical system that combines classical match modeling with modern machine learning to identify value bets across Europe's top football leagues.

Layer 1 — Dixon-Coles Poisson Model

The foundation is a bivariate Poisson model based on the Dixon-Coles (1997) framework. It estimates the expected goals for each team by computing attack and defense strength parameters from historical match results, weighted by recency.

Attack / Defense
Team-level strength parameters estimated via maximum likelihood. Updated weekly.
Rho Correction
Dixon-Coles rho parameter corrects for dependency in low-scoring matches (0-0, 1-0, 0-1, 1-1).
Home Advantage
League-specific home factor computed from 3 seasons of data. Adjusts expected goals upward for home teams.
Scoreline Matrix
Full probability matrix (0-0 through 6-6) generates exact market probabilities for all outcomes.

Layer 2 — XGBoost ML Ensemble

A gradient-boosted decision tree (XGBoost) trained on 1,500+ historical match features refines the base model predictions. The ML layer captures non-linear patterns that Poisson misses: form streaks, fixture congestion, Elo momentum, and xG performance trends.

58 features including:
Attack/Defense strengthElo ratingsxG form (5-match)PPDAFixture congestionH2H historyStreak momentumShots formGoals environmentLambda ratioConfidence scoreRho correction
Training samples: 1,541
CV log-loss: 0.604
ML weight: 85%
Retrained: Weekly (Monday)

Data Sources

API
Football-Data.org
Match results, standings, form tables. Primary data source for historical modeling.
xG
Understat
Expected goals (xG), expected goals against (xGA), PPDA per match. 5-match rolling form.
ELO
Club Elo
Elo ratings for team quality assessment. Used for match quality gates and draw filtering.
ODDS
The Odds API
Real-time bookmaker odds from 30+ books. Line shopping for best available odds and EV calculation.
CSV
Football-Data.co.uk
Historical match stats: shots, corners, cards, referee data, Pinnacle closing odds.
WX
Open-Meteo
Weather conditions at match time. Wind, rain, temperature for forward predictions.

Markets & Selection

The V50 strategy targets low-odds, high-probability bets with positive expected value. Each pick must pass multiple filters before qualification.

Home Win
1.20-1.90 odds
Draw
2.80+ odds
Over 1.5
1.20-1.90 odds
Under 3.5
1.20-1.90 odds
Min EV: 3% — only bets with positive expected value
Min Elo: 3050 combined — filters weak matchups
Shin de-vig: True probability extraction from bookmaker odds
Staking: 2.2% of bankroll — compound growth per bet

Leagues Covered

🏴󠁧󠁢󠁥󠁮󠁧󠁿Premier League
🇪🇸La Liga
🇩🇪Bundesliga
🇮🇹Serie A
🇵🇹Primeira Liga

Backtest Results (3 Seasons)

72.9%
Win Rate
+14.3%
Flat ROI
2,203
Total Bets
3
Seasons

V50 strategy backtested on 2023/24, 2024/25, and 2025/26 seasons. All seasons individually profitable.

Start Copy-Betting

Statistical analysis. Not financial advice. Past performance does not guarantee future results.