NBA Moneyline Betting Pipeline

A machine learning pipeline designed to predict NBA moneyline outcomes, size bets using the Kelly criterion, and eventually generate daily betting insights.

This project is currently under development. It brings together historical NBA data, season-level metrics, and real-time betting odds to model and recommend daily bets. While the core components are being built and tested, the structure supports easy iteration and future expansion.

Project Structure

nba-betting/ ├── data/ # Data staging areas
│ ├── raw/ # Raw pulled data (from Kaggle, etc.)
│ ├── interim/ # Cleaned + aligned tables
│ └── processed/ # Feature matrices for modeling
├── src/
│ ├── ingest/ # Game, team stats, odds ingestors
│ ├── features/ # Feature engineering logic
│ ├── models/ # Training & prediction
│ ├── stake/ # Kelly sizing utils
│ └── pipeline.py # Prefect flow for full pipeline
├── configs/ # YAML configuration files
├── notebooks/ # EDA and prototyping
├── tests/ # Unit tests

Key Features (In Progress)

• Historical game ingestion via kagglehub
• Team-level metrics from Basketball Reference
• Real-time odds ingestion using TheOddsAPI
• Feature engineering including rolling stats and market signals
• Model training using XGBoost with cross-validation
• Kelly criterion stake sizing for bankroll strategy
• Pipeline orchestration with Prefect
• Slack integration to notify live picks

Setup & Usage

1. Clone the repository: git clone https://github.com/yourname/nba-betting.git
   cd nba-betting

2. Install dependencies: poetry install

3. Set required environment variables: export KAGGLE_USERNAME=your_username
   export KAGGLE_KEY=your_key
   export ODDS_API_KEY=your_oddsapi_key
   export SLACK_WEBHOOK_URL=your_slack_webhook

4. Run the pipeline: poetry run python src/pipeline.py --run-date $(date +%F)

Sample Output (Development Preview)

This output is representative of the intended functionality:

• Predicted Probabilities: Thunder: 62.5%, Pacers: 37.5%
• Edge over Market: +6.8%
• Recommended Stake (Fractional Kelly): 3.2% of bankroll
• Slack Message Preview:
  Value Pick: Thunder ML -150 (Implied: 60%, Model: 66.8%) – Bet 2.3 units based on post-loss bounce-back trend.

Planned Improvements

• Incorporate player-level data and injury reports
• Build dashboard for model evaluation and backtesting
• Explore alternative modeling techniques (LightGBM, ensemble methods)