Bitcoin Chain Data Analyzer

This project collects multiple on-chain Bitcoin metrics from JSON files, aligns them over time, and uses LightGBM to train a regression model that predicts the market price.

---

🚀 How to Run

1. Create and activate a virtual environment:

python -m venv venv
venv\Scripts\activate

2. Install dependencies:

pip install -r requirements.txt

3. Train the model and generate predictions:

python train.py --data_dir ./data --target market_price --horizon 1 --out_png prediction.png

---

📊 Findings

• We successfully merged 30+ blockchain metrics (e.g., hash rate, difficulty, unique addresses, transaction volume).
• Trained a LightGBM regressor on the aligned dataset.
• Evaluation metrics:

MAE    : ~14,697
RMSE   : ~23,320
MAPE % : ~21.25
R²     : ~0.47

This means the model explains about half of the market price variance.

---

⚠️ Why errors are relatively high?

1. Blockchain-only features – Metrics like transaction fees, hash rate, difficulty, addresses, etc. capture fundamentals, but price is also influenced by macroeconomics, speculation, and regulations.
2. Non-stationarity – Correlations change between bull and bear markets.
3. Lagging indicators – Many on-chain signals follow the price instead of leading it.

---

📈 Prediction Result

Here’s an example of actual vs predicted prices:

---

✅ Bottom Line

• Blockchain metrics do provide signal (R² ≈ 0.47).
• However, they are not sufficient alone for precise price prediction.
• Improvements could come from:
  • Adding macro and sentiment data (S&P 500, dollar index, Google Trends, Twitter/Reddit).
  • Predicting returns instead of raw prices.
  • Using sequence models (LSTM, GRU, Transformers) to capture temporal dependencies.

---