Crypto Direction Predictor

A real-time ML system that predicts whether Bitcoin price will go UP or DOWN in the next 30 seconds.

System Architecture

CoinGecko API (live price)
        ↓
store_data.py (every 30 sec)
        ↓
PostgreSQL Database
        ↓
retrain.py (every 1 hour)
        ↓
direction_model.pkl
        ↓
FastAPI /predict endpoint
        ↓
Streamlit Dashboard

Tech Stack

• Data Pipeline — Python, CoinGecko API, psycopg2
• Database — PostgreSQL
• ML Model — RandomForestClassifier (scikit-learn)
• Experiment Tracking — MLflow
• API — FastAPI
• Frontend — Streamlit, Plotly
• Environment — Python venv

Features

• Live data collection every 30 seconds
• Auto-retraining every hour on fresh data
• MLflow experiment tracking across runs
• REST API for predictions
• Interactive Streamlit dashboard

Model Performance

Metric	Value
Accuracy	81%
Macro F1	0.77
UP Recall	85%
Naive Baseline	74.7%

Project Structure

crypto_project/
    store_data.py      — live data collection pipeline
    main.py            — FastAPI prediction endpoint
    retrain.py         — automated model retraining with MLflow
    app.py             — Streamlit frontend dashboard
    analysis.ipynb     — EDA and model development
    requirements.txt   — dependencies
    .env               — credentials (not pushed)

📸 Demo

Setup

1. Clone the repo
2. Create virtual environment and install dependencies

python -m venv venv
venv\Scripts\activate
pip install -r requirements.txt

3. Create .env file with your PostgreSQL credentials
4. Create the database table

CREATE TABLE crypto_prices (
    id SERIAL PRIMARY KEY,
    name TEXT NOT NULL,
    price DOUBLE PRECISION NOT NULL,
    timestamp TIMESTAMP DEFAULT CURRENT_TIMESTAMP
);

5. Run the system

# Terminal 1 — collect data
python store_data.py

# Terminal 2 — serve predictions
uvicorn main:app --reload

# Terminal 3 — auto retrain
python retrain.py

# Terminal 4 — dashboard
streamlit run app.py

API Usage

GET http://localhost:8000/predict

Response:

{
  "prediction": "UP",
  "confidence": 54.58,
  "current_price": 70816.0
}

Author

Aditya Kumar — GitHub | LinkedIn