🏀 BucketProps

BucketProps is a data-driven NBA player props platform that uses machine learning to generate over/under point predictions for daily games. The goal is simple: turn historical NBA data into clear, confidence-weighted insights that focus on the numbers to make predictions.

---

🚀 What BucketProps Does

• Predicts NBA player points using machine learning
• Outputs Over / Under picks with confidence scores
• Updates player data continuously via cached NBA stats
• Separates data updates from model retraining for efficiency
• Displays picks in a clean Next.js frontend

---

🧠 How It Works

1. Data Collection

• Historical NBA player game logs
• Vegas lines from prop platforms (e.g. PrizePicks-style props)
• Player-level stats cached locally to avoid repeated API hits

2. Feature Engineering

Features include (but aren’t limited to):

• Rolling & expanding averages
• Recent game performance trends
• Usage-based indicators
• Opponent-adjusted context (where available)

3. Model Training

• Machine learning model trained across multiple seasons
• Designed to generalize across players, not overfit to single games
• Retrained periodically (not daily) to stay aligned with league trends
• Utilizes Gradient Boosting Regression to make accurate predictions

4. Prediction Output

Each pick includes:

• Player name
• Vegas line
• Model-predicted points
• Over / Under recommendation
• Confidence score

---

🖥️ Tech Stack

Backend / ML

• Python
• pandas, NumPy
• scikit-learn
• NBA stats APIs
• Local player cache system

Frontend

• Next.js
• React
• Client-side data fetching from picks.json
• Simple, fast, and mobile-friendly UI

---

📁 Project Structure (Simplified)

bucketprops/
│
├── scripts/
│   ├── update_player_cache.py
│   ├── train_model.py
│   └── predict.py
│
├── data/
│   ├── player_cache/
│   ├── training/
│   └── picks.json
│
├── model/
│   └── .pkl file
│   └── metadata
|
├── frontend/
│   └── Next.js app
│
└── README.md

---

🔁 Model Update Strategy

• Player stats → updated frequently (daily or near-daily)
• Model retraining → done periodically (e.g. bi-weekly)
• This avoids noise from single games while keeping predictions relevant

---

⚠️ Disclaimer

BucketProps is for educational and informational purposes only. Predictions are not guarantees. No financial or betting advice is provided.

---

👤 Author

• Built by Guritfak Gill
• Computer Science student at UC Davis & ML-focused software engineer,
• Passionate about sports analytics, data systems, and machine learning models

---