Note: This dataset is synthetic / for educational and ML-practice purposes. Pre-match features and match outcomes are simulated to model realistic distributions and are not official FIFA results. Do not use it for betting or as a source of truth.

Dataset Overview

A tabular dataset for predicting the outcome of FIFA World Cup 2026 matches. Each row is a single match with engineered pre-match features (rankings, Elo, recent form, head-to-head, squad value) and a full set of post-match target columns, so you can train models for several prediction tasks at once:

Exact score — home_score, away_score
Goal difference & winning margin — goal_difference, winning_margin
Match result — result (home win / draw / away win)
Total goals / Over-Under — total_goals, over_2_5_goals
Both teams to score (BTTS) — both_teams_scored
Goalscorers — first_goalscorer, goalscorers

The tournament covered: 48 teams, 104 matches, hosted across 16 venues in the USA, Canada, and Mexico (June–July 2026).

Total examples: 14,996 matches
- Train (historical international matches, 2006–2026): 14,892 (99.3%)
- Test (the 104 World Cup 2026 fixtures): 104 (0.7%)
Total size: 5.72 MB
Language: English (team / venue metadata)

Dataset Statistics

Split Information

Split	Description	Examples	Size (MB)
Train	Historical internationals (friendlies, qualifiers, past tournaments)	14,892	5.68
Test	FIFA World Cup 2026 fixtures	104	0.04
Total		14,996	5.72

Notes on the Split

Train holds completed historical matches used to learn patterns.
Test holds the 104 World Cup 2026 fixtures; use it as your evaluation/holdout set or for generating tournament predictions.
Random Seed: 42 (for any further re-splitting / reproducibility).

Features

Input features (known before kickoff)

Feature	Type	Description
`match_id`	int64	Unique match identifier
`match_date`	string	Match date (`YYYY-MM-DD`)
`stage`	string	Group Stage, Round of 32, Round of 16, Quarter-final, Semi-final, Third-place, Final
`group`	string	Group letter (A–L) or `Knockout`
`venue`	string	Stadium name
`host_city`	string	Host city
`host_country`	string	USA, Canada, or Mexico
`home_team` / `away_team`	string	National teams
`home_confederation` / `away_confederation`	string	UEFA, CONMEBOL, CONCACAF, CAF, AFC, OFC
`home_fifa_rank` / `away_fifa_rank`	int64	FIFA world ranking at match time
`home_elo` / `away_elo`	float64	Elo rating at match time
`home_form_last5` / `away_form_last5`	string	Last 5 results, e.g. `"WWDLW"`
`home_goals_scored_avg` / `away_goals_scored_avg`	float64	Avg goals scored (last 10 matches)
`home_goals_conceded_avg` / `away_goals_conceded_avg`	float64	Avg goals conceded (last 10)
`h2h_matches`	int64	Total prior head-to-head meetings
`h2h_home_wins` / `h2h_draws` / `h2h_away_wins`	int64	Head-to-head record
`home_squad_value_m` / `away_squad_value_m`	float64	Squad market value (million EUR)
`is_neutral_venue`	bool	Whether neither team is playing at home

Target columns (what you predict)

Feature	Type	Description
`home_score`	int64	Goals scored by home team
`away_score`	int64	Goals scored by away team
`total_goals`	int64	`home_score + away_score`
`goal_difference`	int64	`home_score - away_score` (signed)
`winning_margin`	int64	Absolute goal difference
`result`	string	`home_win`, `draw`, or `away_win`
`outcome_code`	int64	`1` home win, `0` draw, `2` away win
`both_teams_scored`	bool	BTTS flag
`over_2_5_goals`	bool	`total_goals > 2.5`
`first_goalscorer`	string	Player who scored first (empty if `0–0`)
`goalscorers`	string	JSON list, e.g. `[{"player": "...", "team": "...", "minute": 23}]`

Data Format

A single row (illustrative):

{
  "match_id": 14901,
  "match_date": "2026-06-11",
  "stage": "Group Stage",
  "group": "A",
  "venue": "Estadio Azteca",
  "host_city": "Mexico City",
  "host_country": "Mexico",
  "home_team": "Mexico",
  "away_team": "Poland",
  "home_confederation": "CONCACAF",
  "away_confederation": "UEFA",
  "home_fifa_rank": 14,
  "away_fifa_rank": 31,
  "home_elo": 1812.4,
  "away_elo": 1743.9,
  "home_form_last5": "WWDWL",
  "away_form_last5": "DWLDW",
  "home_goals_scored_avg": 1.9,
  "away_goals_scored_avg": 1.3,
  "home_goals_conceded_avg": 0.8,
  "away_goals_conceded_avg": 1.1,
  "h2h_matches": 7,
  "h2h_home_wins": 3,
  "h2h_draws": 2,
  "h2h_away_wins": 2,
  "home_squad_value_m": 312.5,
  "away_squad_value_m": 268.0,
  "is_neutral_venue": false,
  "home_score": 2,
  "away_score": 1,
  "total_goals": 3,
  "goal_difference": 1,
  "winning_margin": 1,
  "result": "home_win",
  "outcome_code": 1,
  "both_teams_scored": true,
  "over_2_5_goals": true,
  "first_goalscorer": "Player A",
  "goalscorers": "[{\"player\": \"Player A\", \"team\": \"Mexico\", \"minute\": 23}, {\"player\": \"Player B\", \"team\": \"Poland\", \"minute\": 58}, {\"player\": \"Player C\", \"team\": \"Mexico\", \"minute\": 77}]"
}

Usage Examples

Load the Dataset

from datasets import load_dataset

dataset = load_dataset("vichetkao/fifa_worldcup_2026_prediction_dataset")

train = dataset["train"]
fixtures = dataset["test"]

print(f"Training matches:        {len(train)}")
print(f"World Cup 2026 fixtures: {len(fixtures)}")

Load as pandas

import pandas as pd

splits = {
    "train": "data/worldcup_2026_train.parquet",
    "test":  "data/worldcup_2026_fixtures.parquet",
}
base = "hf://datasets/vichetkao/fifa_worldcup_2026_prediction_dataset/"

df_train = pd.read_parquet(base + splits["train"])
df_test  = pd.read_parquet(base + splits["test"])

Parse the goalscorers column

import json

row = df_train.iloc[0]
scorers = json.loads(row["goalscorers"])
for g in scorers:
    print(f"{g['minute']}'  {g['player']} ({g['team']})")

Baseline 1 — Predict the result (classification)

import pandas as pd
from sklearn.ensemble import HistGradientBoostingClassifier
from sklearn.metrics import accuracy_score

feature_cols = [
    "home_fifa_rank", "away_fifa_rank", "home_elo", "away_elo",
    "home_goals_scored_avg", "away_goals_scored_avg",
    "home_goals_conceded_avg", "away_goals_conceded_avg",
    "h2h_home_wins", "h2h_draws", "h2h_away_wins",
    "home_squad_value_m", "away_squad_value_m", "is_neutral_venue",
]

X_train, y_train = df_train[feature_cols], df_train["outcome_code"]
X_test,  y_test  = df_test[feature_cols],  df_test["outcome_code"]

clf = HistGradientBoostingClassifier().fit(X_train, y_train)
preds = clf.predict(X_test)
print("Result accuracy:", accuracy_score(y_test, preds))

Baseline 2 — Predict the score (regression)

from sklearn.ensemble import HistGradientBoostingRegressor
from sklearn.metrics import mean_absolute_error

home_model = HistGradientBoostingRegressor().fit(df_train[feature_cols], df_train["home_score"])
away_model = HistGradientBoostingRegressor().fit(df_train[feature_cols], df_train["away_score"])

pred_home = home_model.predict(df_test[feature_cols]).round().clip(0)
pred_away = away_model.predict(df_test[feature_cols]).round().clip(0)

print("Home-goals MAE:", mean_absolute_error(df_test["home_score"], pred_home))
print("Away-goals MAE:", mean_absolute_error(df_test["away_score"], pred_away))

# Derive margin / goal difference / total from the predicted scores
pred_goal_diff = pred_home - pred_away
pred_margin    = abs(pred_goal_diff)
pred_total     = pred_home + pred_away

Baseline 3 — Poisson goal model (margin & exact-score distribution)

import numpy as np
from scipy.stats import poisson

# Expected goals from your regression models above
lam_home = home_model.predict(df_test[feature_cols]).clip(0.1)
lam_away = away_model.predict(df_test[feature_cols]).clip(0.1)

def scoreline_matrix(lh, la, max_goals=6):
    h = poisson.pmf(np.arange(max_goals + 1), lh)
    a = poisson.pmf(np.arange(max_goals + 1), la)
    return np.outer(h, a)   # P(home=i, away=j)

m = scoreline_matrix(lam_home[0], lam_away[0])
home_win = np.tril(m, -1).sum()
draw     = np.trace(m)
away_win = np.triu(m, 1).sum()
print(f"P(home win)={home_win:.2f}  P(draw)={draw:.2f}  P(away win)={away_win:.2f}")

Suggested Tasks

Task	Target column(s)	Type
Match result (1X2)	`result` / `outcome_code`	Classification
Exact score	`home_score`, `away_score`	Regression / Poisson
Goal difference & margin	`goal_difference`, `winning_margin`	Regression
Total goals / Over-Under 2.5	`total_goals`, `over_2_5_goals`	Regression / Classification
Both teams to score	`both_teams_scored`	Classification
First goalscorer	`first_goalscorer`	Multi-class

File Summary

File	Type	Size	Samples
worldcup_2026_train.parquet	Parquet	5.68 MB	14,892
worldcup_2026_fixtures.parquet	Parquet	0.04 MB	104

Citation

@dataset{vichetkao_worldcup2026_prediction_2026,
  title  = {FIFA World Cup 2026 Match Prediction Dataset},
  author = {Vichet Kao},
  year   = {2026},
  url    = {https://huggingface.co/datasets/vichetkao/fifa_worldcup_2026_prediction_dataset},
  note   = {Synthetic tabular dataset for football match outcome, score, margin, and goalscorer prediction}
}

License

CC-BY-4.0

Disclaimer

Data is synthetic and for educational / modeling-practice use only. It does not represent official FIFA records and must not be used for gambling or as factual match data.

Contact & Support

Open a discussion on the dataset repository.

Last Updated: 2026-06-17 Dataset Version: 1.0 Total Examples: 14,996 Total Size: 5.72 MB Tasks: Score / result / margin / goal-difference / goalscorer prediction

Downloads last month: 5

Total file size:

1.32 GB