I engineered 102 leakage-free ML features from 49,000+ international football matches (1872–2026) and published it as a free dataset

Been working on a football prediction project and couldn't find a dataset that had the actual context needed to model match outcomes — just raw results everywhere. So I built one from scratch on top of the International Football Results dataset by Mart Jürisoo (the well known one on Kaggle with 49,000+ matches going back to 1872). What I added: **Elo ratings** — built from scratch, updated after every single match across 150 years. Both teams' ratings, their difference, and the expected win probability going into each match. **Rolling form** — win rate, goals scored, goals conceded, goal diffe