NBA game outcome model
Machine-learning coursework
A historical-data pipeline joins NBA team statistics and game totals, then classifies a combined game outcome.
- Status
- Coursework prototype
- Kind
- Data & geometry
- Stack
- Python · pandas · scikit-learn · Random forests
BUILD NOTES / Coursework prototypeNBA game outcome model
Project overview · not an interface capture
Machine-learning coursework
Python / pandas / scikit-learnProject overview · not an interface capture
The problem
The model needed consistent game identities and historical features before it could compare predictions against held-out outcomes.
What was built
- Cleaning and joining historical game records with total lines by game identifier.
- Lagged five-game team features and a chronological training/test split.
- A 100-tree random forest for the combined home-win and over-total label, with model and feature-list serialization.
What I learned
The prediction target, feature timing, and evaluation split need to be explicit. A successful training run does not establish useful real-world predictions.