🏀 NBA Game Outcome Predictor

Predicts whether the home team wins an NBA game from a minimal set of box‑score stats.

Tech stack: Python · Pandas · Scikit‑learn · Google Colab

At a glance

Data	26 000+ games (2004 – 2020) from public box-score archives
Pre-processing	Full-row NaN purge · automatic binary Win label generation
Features	FG % (home & away) · rebounds (home) · assists (home)
Model	Random Forest, 100 trees, max_depth=None
Validation	Stratified 20 % hold-out · 79 % accuracy
Diagnostics	Confusion-matrix heat map to verify class balance
Extras	Match-up Simulator – plug your stats ⇒ instant W/L call

Quick start (Colab)

Click the badge or run:

!git clone https://github.com/dcheongsee/NBA‑Predictor.git

Open NBA_Predictor.ipynb in Colab and execute Run All – the notebook rebuilds the entire pipeline end‑to‑end in under two minutes on free tier hardware.

Local run

python -m venv .venv && source .venv/bin/activate
pip install pandas scikit-learn matplotlib notebook
jupyter notebook NBA_Predictor.ipynb

Repository layout

.
├── NBA_Predictor.ipynb   # Reproducible workflow
|                         # CLI + Match‑up simulator
├── nba_games.csv         # cached CSV
└── README.md

Results

• Accuracy   79 %
• F1‑score   0.78

See NBA_Predictor.ipynb for full metrics.

Why it matters

A compact but end‑to‑end example of sports analytics: raw CSV ingestion through cleaning, feature engineering, modelling and live inference. Reproducible in a single notebook.

License

MIT-licensed. Keep the copyright header, otherwise do whatever you like