Data Shop › Football › Football — Dataset Machine Learning · 9…
Football — Dataset Machine Learning · 96 stats pré-match (2021-2027)
Pour entraîner vos propres modèles de prédiction : 27 190 matchs avec, pour chacun, 96 statistiques calculées AVANT le coup d'envoi (forme des équipes sur 5/10/20 matchs, buts attendus xG, jours de repos, classement, tirs, corners, cartons…) et le résultat réel à prédire. Garantie sans triche : chaque stat n'utilise que l'information disponible avant le match — vérifié sur 150 matchs recalculés avec coupure stricte, 138 recalculables et 138/138 conformes. Se joint aux datasets de cotes par la clé match_id.
What this dataset contains
- 96 statistiques pré-match par ligne, calculées par ProbWin
- OAxG exclusif : xG ajusté à la force des adversaires rencontrés (7 colonnes)
- Sans fuite de futur : chaque stat n'utilise que l'info d'avant-match (vérifié par recalcul)
- Cibles incluses : résultat 1X2 et total de buts réels
- 27 190 matchs · 15 championnats · saisons 2021-22 à 2026-27
- Se joint aux datasets 1X2 / Over-Under par match_id
- CSV + Parquet · dictionnaire des 96 colonnes inclus
Stat families 96 columns, detailed in the README
| Family | Content |
|---|---|
| To predict | Match result (win/draw/loss) and goal total — computed from real scores |
| Goal form | Goals scored/conceded over the last 5, 10 and 20 matches, home and away |
| xG (expected goals) | Quality of the chances each team creates and concedes |
| OAxG — ProbWin exclusive | Our opponent-adjusted xG: xG inflated against weak defences is corrected down, xG earned against strong defences is upgraded (7 columns: for/against/differential per team + combined) |
| Form & schedule | Points over 5/10 matches, average margin, rest days, match density over 14 days |
| Standings | League rank, position gap between the two teams, cumulative goal difference |
| Shots, corners, cards | Shots on target for/against, ratios, corners over 10 matches, weighted cards (yellow = 1, red = 3) |
| League context | Average goals and share of over-2.5 matches in the league |
Leak-free guarantee ("point-in-time"): every stat uses only information available BEFORE the match. Verified on a random sample of 150 matches recomputed with a strict cutoff: 138 were recomputable (the other 12, at the very start of a season, lack the required history) — 138/138 conformant, no future leakage.
Real file preview 3 rows taken from the file you buy · + 88 autres colonnes
How to read it: Brentford host Arsenal with an OAxG of 2.70 vs 1.94 — Brentford's attack over-produces even after accounting for the defences it faced (goals_l5 = average goals scored over the last 5 matches; rest_days = days of rest). All 96 columns are listed in the README.
Coverage per season
| Season | Matches with all 96 stats |
|---|---|
| 2021-22 | 5 436 |
| 2022-23 | 5 394 |
| 2023-24 | 5 368 |
| 2024-25 | 5 250 |
| 2025-26 | 5 229 |
| 2026-27 | 513 (en cours) |
Frequently asked questions
Which leagues are included?
Premier League, Championship, LaLiga, LaLiga2, Bundesliga, 2. Bundesliga, Serie A, Serie B, Ligue 1, Ligue 2, Eredivisie, Liga Portugal, Jupiler Pro League, Süper Lig, Super League — both top divisions of the big five European countries plus 5 major leagues, each present in all 6 seasons.
What happens after payment?
You immediately receive a secure download link (5 downloads). The archive contains the CSV and Parquet files plus the README documenting every column and its exact coverage, season by season.
Can I join it with the other datasets?
Yes, all our football datasets share the match_id key.
Why 27,190 matches here vs 27,109 on the odds side?
The 96 stats can be computed for any played match; the odds datasets additionally
require archived odds. 81 matches have their stats but no usable odds —
they are here, not there. Joining on match_id aligns the rest.
What if the file does not suit me?
Digital product: no refund after download. That is why everything is verifiable before purchase — real preview, columns one by one, quantified per-season coverage — and we answer any pre-purchase question at [email protected].
Is the file updated after purchase?
No — you buy the file as described on this page, at its stated cut-off date. Finished seasons are final. To follow the season being played, there is the Season in Progress product (€12, weekly updates included).
A question before buying?
[email protected] — we answer personally.
One-time payment — no subscription
Buy — instant download- Secure Stripe payment
- Instant download link
- CSV + Parquet + quantified README
- Link valid 72 h · 5 downloads
- Single-user licence — your models and analyses (commercial included) are yours; resale of the files prohibited
- Support: [email protected]
A free ProbWin account is required at checkout (it holds your download links).
Free sample (50 rows, all columns)Complete it for backtesting ROI is measured against odds — join on match_id
Or take the Complete Football Pack — this dataset + both odds datasets for €69 instead of €97.
