A sports prediction engine where every ML algorithm is built from scratch.
I'm a software engineer learning machine learning by implementing it. PredictaKit predicts match outcomes using algorithms I write by hand with NumPy, and every prediction explains why it made that call.
Follow along via commits and the learning journal.
Point it at historical match data (football, cricket, basketball), and it predicts outcomes with confidence scores and factor breakdowns. The engine (PredictaCore, living in core/) is the from-scratch ML — linear regression, logistic regression, decision trees, ensembles. The toolkit wraps it into something usable.
- Project setup
- Linear regression from scratch
- Logistic regression from scratch
- Decision trees from scratch
- Ensemble + explanation engine
- Multi-sport features, CLI, ship it
- No sklearn, PyTorch, or TensorFlow in the engine. NumPy only.
- Every prediction must be explainable.
- I write every algorithm. AI helps with scaffolding and tests, not the math.
https://www.football-data.co.uk/englandm.php
MIT