AI Football Match Winner Predictor

I need a production-ready machine-learning system that reliably predicts the match winner (home win, draw, or away win) for virtually any football game as long as the raw data exists. The model must ingest every relevant signal you can legally obtain—historic results, team and player form, injury reports, betting odds, even live in-play stats when they are available—and turn it into a single probability distribution for each possible outcome.

Here is how I picture the workflow:

• Data pipeline: automated scripts that collect, clean, and update datasets from multiple public free sources.
• Feature engineering & modelling: state-of-the-art algorithms (e.g., gradient boosting, XGBoost, deep learning, ensemble stacking) with clear justification for each choice, plus hyper-parameter tuning.
• Evaluation: back-testing across seasons and leagues, reporting accuracy, log-loss, and profit-based metrics against bookmaker odds.
• Deployment: Jupiter notebook

Acceptance criteria
1. End-to-end script runs automatically and reproduces published metrics.
2. Model achieves a statistically significant edge over baseline bookmaker odds (at least 60% real accuracy)

If you have experience blending sports analytics with machine learning, let’s talk about the approach you’d follow and the datasets or services you already use successfully.