Skip to content

Honest numbers.

Scored on 297 matches it never saw in training. Lower log-loss is better.

Performance over timelog-loss, better sits higher ↑
Matchday FateXGBoost aloneBet365
0.9450.9801.01520/2121/2222/2323/2424/2525/2620/21, Bet365: 0.982521/22, Bet365: 0.977222/23, Bet365: 0.978923/24, Bet365: 0.954824/25, Bet365: 0.964725/26, Bet365: 0.979020/21, XGBoost alone: 1.000321/22, XGBoost alone: 0.997922/23, XGBoost alone: 0.990023/24, XGBoost alone: 0.983024/25, XGBoost alone: 0.985325/26, XGBoost alone: 0.995720/21, Matchday Fate: 0.997321/22, Matchday Fate: 0.993822/23, Matchday Fate: 0.988823/24, Matchday Fate: 0.978424/25, Matchday Fate: 0.983925/26, Matchday Fate: 0.9923

Six seasons, each predicted only from earlier seasons. The blend beats XGBoost alone every season; Bet365 stays ahead by about 0.02.

23

pre-match signals

XGBoost + logistic

blended, then calibrated

Strict time order

never trained on the future

Beats naive

trails Bet365 by ~0.02 log-loss

Log-loss, lower is better

Naive guess

always the training average

1.0674

XGBoost alone

23 fair-play signals

0.9613

Matchday Fate

XGBoost + logistic blend, calibrated

0.9596

Bookmaker

Bet365, the benchmark

0.9698

On this small window the model edges the bookmaker, but that gap is inside the noise. Over full seasons the bookmaker is still ahead by about 0.02 — expected, since it prices in team news this model never sees. Every change to the model has to beat the old one in all six past seasons before it ships.

What goes in

  • Elo strength (plain + margin-adjusted)
  • Form over last 5 and 10
  • Goals, shots, shots on target, corners
  • Rest days
  • Head-to-head streak

What stays out

  • Betting odds of any kind
  • Anything known only after kickoff
  • Lineups and injuries (no free source yet)

Research demo, not betting advice.