⚽Football Predictor

📊 Model Accuracy

Live accuracy tracking across all completed matches with ML predictions. Refreshed every hour · last computed 29 Jul 2026, 12:41.

⚠️ Model change — 2026-06-17

The model became market-independent on 2026-06-17 (market features + anchoring removed). The 498 games before then were served by the old (anchored) model, while 281 by the current one. The “All Time” numbers below mix the two methodologies — the rolling 7d/30d are more representative of the current model.

Model period1×2O/UN
anchored (… → 2026-06-17)51.0%57.6%498
pure-model (2026-06-17 → 2026-07-10)68.8%56.3%80
pure-unified (2026-07-10 → now)48.3%55.7%201
⚡

Top AI Picks — historical accuracy

Top 3 per day · high confidence → higher probability

Accuracy

66%

170 / 256 correct

Vs Overall

+14%

diff from 52% overall

Avg Probability

66%

top pick confidence

Total Picks

256

~3 per match day

1×2 Result

70%

98 / 140 correct

140 picks

Over/Under 2.5

62%

72 / 116 correct

116 picks

Same logic as the home page: top 3 matches per day ranked by high confidence → highest outcome probability. Counts whether the predicted outcome (Home Win / Away Win / Draw / Over / Under) was correct.

All Time · 779 matches

Result Accuracy

52%

406 / 779 correct

O/U Accuracy

57%

444 / 779 correct

Both Correct

29%

229 / 779

Matches Tracked

779

with stored predictions

Rolling Performance

Last 7 days78 matches

Result44%
O/U55%
Both22%

Last 30 days228 matches

Result51%
O/U55%
Both28%

By League

LeagueGamesResult %O/U %Both %
🌍International20761%53%33%
🤝Club Friendlies13348%55%26%
🏴󠁧󠁢󠁥󠁮󠁧󠁿Premier League6148%62%33%
🇪🇸La Liga5648%61%25%
🇮🇹Serie A5553%55%24%
🇫🇷Ligue 14646%63%30%
🇩🇪Bundesliga4245%69%36%
🇬🇷Super League4250%57%24%
🟢Conference League3549%60%31%
⭐Champions League2143%67%33%
🇳🇱Eredivisie2045%55%25%
🇵🇹Primeira Liga1958%53%32%
🇧🇷Brasileirão1644%44%19%
🟠Europa League1560%60%40%
🏴󠁧󠁢󠁥󠁮󠁧󠁿Championship1164%55%27%

🌍 International — By Tournament

LeagueGamesResult %O/U %Both %
🌍AFC Asian Cup5159%55%25%
🌍AFC Asian Cup qualification6568%48%32%
🌍ASEAN Championship2658%62%42%
🌍ASEAN Championship qualification20%50%0%
🌍African Cup of Nations10454%59%25%
🌍African Cup of Nations qualification15154%59%29%
🌍Al Ain International Cup425%75%25%
🌍Arab Cup3161%45%19%
🌍Arab Cup qualification743%14%0%
🌍Baltic Cup667%83%50%
🌍CONCACAF Nations League11464%54%39%
🌍CONCACAF Series3546%66%29%
🌍CONIFA Asia Cup4100%100%100%
🌍Canadian Shield450%75%25%
🌍Copa América3256%53%25%
🌍Copa América qualification250%0%0%
🌍Diamond Jubilee International Football Tournament250%100%50%
🌍EAFF Championship qualification580%80%80%
🌍FIFA Series5763%54%33%
🌍FIFA World Cup10666%51%34%
🌍FIFA World Cup qualification76365%58%36%
🌍Friendly65858%57%30%
🌍Gold Cup3165%61%35%
🌍Gold Cup qualification1471%71%43%
🌍Gulf Cup1527%47%13%
🌍Intercontinental Cup367%33%0%
🌍Island Games1560%87%60%
🌍King's Cup875%38%38%
🌍Kirin Cup450%0%0%
🌍MSG Prime Minister's Cup1136%36%27%
🌍Mapinduzi Cup743%86%43%
🌍Marianas Cup250%100%50%
🌍Merdeka Tournament475%75%50%
🌍Morocco, Capital of African Football617%67%17%
🌍Mukuru 4 Nations250%50%50%
🌍Oceania Nations Cup1369%46%31%
🌍Oceania Nations Cup qualification367%33%33%
🌍Outrigger Challenge Cup30%100%0%
🌍Soccer Ashes2100%50%50%
🌍South Asian Super Cup10%0%0%
🌍Tri-Nations Cup1100%0%0%
🌍Tri-Nations Series333%67%33%
🌍UEFA Euro5149%59%25%
🌍UEFA Euro qualification978%33%22%
🌍UEFA Nations League18659%51%28%
🌍Unity Cup875%25%25%

By Confidence Level (club leagues)

🟢 high109 matches

Result accuracy61%

66 / 109 correct

🟡 medium175 matches

Result accuracy46%

80 / 175 correct

🔴 low288 matches

Result accuracy46%

133 / 288 correct

High confidence = max outcome probability ≥ 55% AND signal on O/U · Medium ≥ 42% · Low < 42%

Internationals (separate scale: HIGH = p ≥ 65%)

🟢 high62 matches

Result accuracy82%

51 / 62 correct

🟡 medium40 matches

Result accuracy65%

26 / 40 correct

🔴 low105 matches

Result accuracy48%

50 / 105 correct

By Predicted Outcome

Match Result

🏠 Home win55%

283 / 513 correct

🤝 Draw42%

14 / 33 correct

✈️ Away win47%

109 / 233 correct

Over / Under 2.5

⬆️ Over 2.559%

281 / 477 correct

⬇️ Under 2.554%

163 / 302 correct

Draw Prediction

Total Draws

198

actual draws in dataset

Draw Predictions

33

matches predicted as draw

Draw Recall

7%

of actual draws caught

Draw Precision

42%

of draw predictions correct

Draws (25% of matches) are the hardest outcome to predict. Recall = what fraction of actual draws we caught · Precision = how reliable our draw calls are.

Goal / No Goal (BTTS)

GG (Goal Goal) = both teams scored at least 1 goal. NG (No Goal) = at least one team didn't score. The prediction is made via the Poisson model (from the stored λ).

Sample

640

completed matches with λ

337 GG/303 NG

Overall Accuracy

50%

323 / 640 correct (GG+NG)

GG Accuracy

52%

220 / 420 GG predictions correct

GG Recall

65%

Of 337 actual GG, we caught 220

NG Recall

34%

Of 303 actual NG, we caught 103

GG Predictions

420

matches we predicted GG

NG Predictions

220

matches we predicted NG

ROI Tracker

💰 ROI Tracker — Value Strategy

Only ⚡ suggested bets · €10 flat · 221 bets

€-474.50

Strategy ROI -21.47%

📉 Closing Line Value

16 bets with closing snapshot · positive CLV = real edge

-7.79%

beat close 31%

🎯 Fair-value ROI — no vig

Same bets at fair (de-vigged) odds · model quality vs market

+€120.73

+1.24% fair

vs −€309.30 (-3.17%) with vig

Model baseline (bet everything · 977 bets)

−€309.30 · -3.17%

+€120.73 · +1.24%

with vigmodel (fair)

1×2 Result

466 bets · €4660.00 staked · vig −€226.22

−€106.20

-2.28%

+€120.02

+2.58% fair

Over 2.5 Goals

319 bets · €3190.00 staked · vig −€122.34

−€131.40

-4.12%

−€9.06

-0.28% fair*

GG (BTTS)

192 bets · €1920.00 staked · vig −€81.46

−€71.70

-3.73%

+€9.76

+0.51% fair

Real result (with vig)−€309.30
↳ from correct model results+€120.73
↳ lost to vig (bookmaker margin)−€430.03

Strategy = flat stake only on ⚡ suggested value bets (with market-shrunk EV gate). The baseline bets on every prediction and is expected ≈ −vig — it's a model-health signal, not a strategy. Fair-value = same bets at de-vigged odds (Result & BTTS exactly; *O/U with a hypothetical 4% overround since we don't store under-2.5 odds). It's not an achievable return — nowhere can you bet at fair odds — but it cleanly measures model quality.

Cumulative EV vs P&L

📈 Cumulative EV vs P&L

€10 flat stake · 76 days tracked

Expected ValueActual P&LFair P&L (no vig)
-€600-€400-€200+€0+€200+€40003-07 04-30 05-13 06-11 06-24 07-09 07-26

Calibration

O/U Calibration — predicted vs actual over-rate

Points near the diagonal = well calibrated. Above = model over-predicts, below = under-predicts.

Predicted: 35% | Actual: 36% | n=25Predicted: 40% | Actual: 41% | n=158Predicted: 48% | Actual: 58% | n=144Predicted: 55% | Actual: 54% | n=151Predicted: 60% | Actual: 61% | n=195Predicted: 68% | Actual: 61% | n=69Predicted: 76% | Actual: 61% | n=23Predicted: 81% | Actual: 83% | n=6Predicted: 87% | Actual: 100% | n=20%25%50%75%100%30%50%70%90%Predicted Over ProbabilityActual Over Rate
Perfect calibrationModel· bubble size = sample count

1×2 Result Calibration — predicted vs actual frequency

Points on the diagonal = well calibrated. Above = model under-estimates, below = over-estimates. Minimum 3 matches per bucket shown.

Home: predicted 7% | actual 5% | n=21Home: predicted 16% | actual 20% | n=59Home: predicted 26% | actual 23% | n=84Home: predicted 35% | actual 38% | n=144Home: predicted 45% | actual 53% | n=169Home: predicted 54% | actual 48% | n=168Home: predicted 64% | actual 67% | n=64Home: predicted 74% | actual 73% | n=44Home: predicted 83% | actual 71% | n=21Home: predicted 92% | actual 80% | n=5Draw: predicted 7% | actual 25% | n=8Draw: predicted 16% | actual 25% | n=95Draw: predicted 25% | actual 24% | n=537Draw: predicted 33% | actual 29% | n=117Draw: predicted 43% | actual 41% | n=22Away: predicted 6% | actual 8% | n=83Away: predicted 16% | actual 14% | n=115Away: predicted 25% | actual 27% | n=249Away: predicted 35% | actual 30% | n=147Away: predicted 43% | actual 40% | n=80Away: predicted 54% | actual 54% | n=46Away: predicted 63% | actual 60% | n=40Away: predicted 76% | actual 86% | n=14Away: predicted 84% | actual 60% | n=50%25%50%75%100%10%20%30%40%50%60%70%80%Predicted Outcome ProbabilityActual Frequency
Perfect calibration🏠 Home win🤝 Draw✈️ Away win· bubble size = sample count

BTTS Calibration — predicted GG probability vs actual GG rate

Points near the diagonal = well calibrated. Derived from Poisson λ stored at prediction time.

Predicted: 25% | Actual: 40% | n=5Predicted: 36% | Actual: 58% | n=38Predicted: 47% | Actual: 53% | n=175Predicted: 53% | Actual: 51% | n=367Predicted: 64% | Actual: 63% | n=46Predicted: 74% | Actual: 83% | n=60%25%50%75%100%10%30%50%70%90%Predicted BTTS (GG) ProbabilityActual GG Rate
Perfect calibrationModel· bubble size = sample count

By Model Version

VersionGamesResult %O/U %
1.0.057249%59%
national20761%53%