Apppublic

Marchelo23/mempool

sourceHugging Facemitupdated 5mo agoView on Hugging Face

0likes

App README

⚡ Bitcoin Mempool Fee Predictor API

ML-powered real-time fee predictions for Bitcoin block inclusion.

Uses an XGBoost + LightGBM ensemble trained on live mempool data to predict the optimal fee rate (sats/vByte) needed to get your transaction confirmed in the next 1, 3, or 6 blocks.

🔗 API Endpoints

Endpoint	Description
`GET /`	API info & status
`GET /health`	Health check
`GET /fees/predict`	ML fee predictions (1, 3, 6 blocks)
`GET /fees/current`	Current mempool fees (from mempool.space)
`GET /mempool/blocks`	Projected mempool blocks
`GET /models`	Loaded model information
`GET /model-metadata`	Model training metrics & metadata
`GET /docs`	Interactive API docs (Swagger UI)

📡 Example

bash

curl https://marchelo23-mempool.hf.space/fees/predict

Response

json

{
  "timestamp": "2026-04-23T12:00:00",
  "mempool_snapshot": {
    "tx_count": 45000,
    "vsize_mb": 120.5,
    "total_fee_btc": 1.234
  },
  "fee_predictions": {
    "1_block": {
      "predicted_fee_sat_vb": 42,
      "confidence_interval": [36, 48],
      "priority": "high"
    },
    "3_blocks": {
      "predicted_fee_sat_vb": 28,
      "priority": "medium"
    },
    "6_blocks": {
      "predicted_fee_sat_vb": 15,
      "priority": "low"
    }
  },
  "recommendation": "NORMAL"
}

🧠 Models

—XGBoost (v2.1+): Primary model, 0.6 ensemble weight
—LightGBM (v4.5+): Secondary model, 0.4 ensemble weight
—Training: Automated retraining on live mempool snapshots
—Features: 100+ engineered features from mempool state, fee landscape, block timing, and network metrics

🏗️ Architecture

mempool.space API → Feature Engineering → Ensemble (XGB + LGB) → Fee Predictions

Built with FastAPI, deployed as a Docker container on Hugging Face Spaces.

📊 Data Source

All data is fetched in real-time from the mempool.space public API (Bitcoin mainnet).