🏀 NBA Oracle

🌐 Live Demo: https://nba-oracle-flyqyf1oh-souppmans-projects.vercel.app

ML-Powered NBA Fantasy Point Projections
Context-Aware. Time-Decayed. Precision Engineered.

---

🎯 Overview

NBA Oracle is a machine learning application that predicts individual NBA player statistics (Points, Rebounds, Assists) for upcoming games. Unlike simple season averages, it uses context-aware features and time-decay weighting to generate accurate daily fantasy projections.

---

🧠 Machine Learning Pipeline

The prediction engine is built on a Multi-Output Regression model trained on ~500,000 historical player game logs.

Key Features:

• Rolling Window Statistics: Calculates mean, standard deviation, and momentum over 3, 5, and 10-game windows
• Opponent Defense Modeling: Incorporates how many points/rebounds the opposing team allows on average
• Fatigue Detection: Tracks rest days and back-to-back game indicators
• Time-Decay Weighting: Recent games are weighted exponentially higher (60-day half-life) to capture current form

Model Architecture:

• Algorithm: XGBoost Multi-Output Regressor
• Targets: Points, Rebounds, Assists (simultaneous prediction)
• Performance:
  • Points: R² = 0.58, RMSE = 5.7
  • Rebounds: R² = 0.51, RMSE = 2.3
  • Assists: R² = 0.55, RMSE = 1.7

The model is trained in a Jupyter Notebook (NBA_PlayerGamePredictor.ipynb) and outputs predictions to CSV files.

---

🔌 UI Integration

The web application is a React + Vite frontend that consumes the ML predictions:

1. Data Layer (src/data/predictions.js): Pre-loaded predictions from the latest CSV export
2. Component Layer (src/components/PlayerCard.jsx): Displays individual player projections with fantasy score calculations
3. App Layer (src/App.jsx): Implements search, filtering, and sorting logic

User Features:

• 🔍 Search: Filter by player name or team
• 📊 Sort: Rank players by Points, Rebounds, or Assists
• ⚡ High Confidence Indicators: Visual highlights for players projected >20 points

---

🛠️ Tech Stack

Machine Learning

• Python 3.x
• Pandas - Data manipulation
• Scikit-learn - Model training & evaluation
• XGBoost - Gradient boosting regressor
• NumPy - Numerical computing

Frontend

• React 19 - UI framework
• Vite - Build tool
• Tailwind CSS v4 - Styling
• Framer Motion - Animations
• Lucide React - Icons

Deployment

• Vercel - Hosting & CI/CD
• PostCSS - CSS processing

---

🚀 Live Demo

https://nba-oracle-flyqyf1oh-souppmans-projects.vercel.app

---

📦 Installation

# Clone the repository
git clone https://github.com/souppman/NBA-Loracle.git
cd NBA-Loracle/nba-oracle

# Install dependencies
npm install

# Run development server
npm run dev

# Build for production
npm run build

---

📊 Data Sources

All data is sourced from publicly available NBA statistics. No proprietary or sensitive information is included in this repository.

---

📄 License

MIT License - See LICENSE file for details

---

🙏 Acknowledgments

Built as a demonstration of applied machine learning in sports analytics. Model training code available in NBA_PlayerGamePredictor.ipynb.