Been working on a football prediction project and couldn’t find a dataset that had <\/p>\n
the actual context needed to model match outcomes \u2014 just raw results everywhere.<\/p>\n
So I built one from scratch on top of the International Football Results dataset <\/p>\n
by Mart J\u00fcrisoo (the well known one on Kaggle with 49,000+ matches going back to 1872).<\/p>\n
What I added:<\/p>\n
**Elo ratings** \u2014 built from scratch, updated after every single match across 150 <\/p>\n
years. Both teams’ ratings, their difference, and the expected win probability <\/p>\n
going into each match.<\/p>\n
**Rolling form** \u2014 win rate, goals scored, goals conceded, goal difference, clean <\/p>\n
sheet rate, both-teams-scored rate, scoring rate, and win streak. Computed at <\/p>\n
three lookback windows: last 5, last 10, and last 20 matches. For both teams.<\/p>\n
**Head-to-head history** \u2014 based on the last 10 meetings between those two specific <\/p>\n
teams. Some teams have persistent edges over specific opponents that their general <\/p>\n
form doesn’t explain.<\/p>\n
**Fatigue signals** \u2014 days since each team’s last match and the difference between <\/p>\n
the two.<\/p>\n
**Penalty reliance** \u2014 fraction of each team’s historical goals that came from <\/p>\n
penalties, pulled from the goalscorer dataset.<\/p>\n
**Shootout composure** \u2014 historical penalty shootout win rate for each team, from <\/p>\n
the shootouts dataset.<\/p>\n
**Tournament context** \u2014 World Cup, qualifier, friendly, neutral venue, competition <\/p>\n
importance weight, confederation.<\/p>\n
The thing I spent the most time on: every feature is computed in strict <\/p>\n
chronological order using only data that existed before that match was played. <\/p>\n
State updates happen after each row is recorded, never before. No lookahead, <\/p>\n
no leakage anywhere in the 102 columns.<\/p>\n
102 features total. 49,094 rows. result column (H\/D\/A) included as the label. <\/p>\n
Drop date and result, plug into any classifier.<\/p>\n
Dataset is fully documented with column descriptors for every feature.<\/p>\n
Link: https:\/\/www.kaggle.com\/datasets\/kriishgulati\/football-match-results-1872-2026-with-ml-features<\/a><\/p>\n Built on top of the original dataset by Mart J\u00fcrisoo \u2014 full credit and link <\/p>\n in the dataset description.<\/p>\n<\/div>\n submitted by \/u\/Kriish_Gulati <\/a>
[link]<\/a><\/span> [comments]<\/a><\/span><\/p>
