NBA Game Predictor

A machine learning project that predicts NBA game outcomes using historical data, ELO ratings, and team statistics. Built with XGBoost and Streamlit.

Live Demo

Access the app: https://borniapedro-nba-game-predictor.streamlit.app/

Overview

This project uses machine learning to predict NBA game winners by analyzing:

• ELO ratings and momentum
• Team fatigue (days between games, back-to-back games)
• Season and rolling statistics (points, FG%, rebounds, turnovers)
• Recent form (last 5 games)

The model is trained on 10,000+ regular season games from 2019 to present and achieves ~64% accuracy.

Installation

git clone https://github.com/BorniaPedro/NBA-game-predictor.git
cd NBA-game-predictor
pip install -r requirements.txt
streamlit run app.py

Usage

Online: Visit the live demo

Locally:

1. Run streamlit run app.py
2. Select a game date
3. Click "Make Predictions"
4. View predictions with confidence levels:
   • 🔥 High Confidence (>65%)
   • ⚖️ Balanced (55-65%)
   • 🎲 Uncertain (<55%)

Project Structure

NBA-game-predictor/
├── app.py                              # Streamlit web app
├── requirements.txt                    # Dependencies
├── notebooks/
│   ├── game_predictor.ipynb           # Model training
│   └── next_game_predictor.ipynb      # Prediction pipeline
├── dataframes/
│   ├── nba_games_2017_today.parquet   # Raw data
│   └── nba_games_history.parquet      # Processed data
└── xgb model/
    ├── nba_model_xgb.pkl              # Trained model
    └── feature_list.pkl               # Features

Technologies

• Streamlit - Web interface
• XGBoost - Machine learning model
• pandas - Data processing
• nba_api - NBA data source
• scikit-learn - Model evaluation

Model Details

• Algorithm: XGBoost Classifier
• Training data: 10,000+ games (2019-present)
• Features: 30+ engineered features
• Accuracy: ~64%
• Data source: NBA API (regular season only)

Legal Notice

This project is for educational purposes only. Not affiliated with the NBA. Does not promote gambling.

Author

Pedro Bornia - @BorniaPedro