Arrythmia-Detection: Federated Learning with MIT-BIH Dataset

Overview

Arrythmia-Detection is a federated learning system built using Flower (flwr) for arrhythmia detection on the MIT-BIH dataset. The project demonstrates privacy-preserving distributed model training, where multiple clients collaboratively train a model without sharing raw data.

Architecture

• Server (server.py)

  • Orchestrates federated learning rounds.
  • Aggregates model updates and evaluation metrics from clients.
  • Uses the FedAvg strategy for model aggregation.
  • Configurable for number of rounds and client requirements.
  • Handles logging and suppresses unnecessary warnings.

• Clients (client1.py, client2.py, client_base.py)

  • Each client trains locally on its own data.
  • Inherit shared logic from client_base.py.
  • Perform local training, evaluation, and send updates/metrics to the server.

• Data Preparation (prep_mitbih.py)

  • Prepares and preprocesses the MIT-BIH arrhythmia dataset for client consumption.

Communication Flows

• Server ↔ Clients

  • Server sends global model and training instructions to clients.
  • Clients send model updates and evaluation metrics to the server.
  • Heartbeat/status signals for monitoring client availability.

• Data Provisioning

  • Clients load and preprocess data using prep_mitbih.py.

Deployment

• The server and clients can be deployed on cloud services, on-premises machines, or containers.
• Secure communication (SSL/TLS) is recommended for production.
• Logging and monitoring are integrated for system health and debugging.

File Structure

Arrythmia-Detection/
│
├── server.py           # Federated server logic and orchestration
├── client_base.py      # Base class for client logic
├── client1.py          # Client 1 implementation
├── client2.py          # Client 2 implementation
├── prep_mitbih.py      # Data preparation for MIT-BIH dataset

Getting Started

Prerequisites

• Python 3.7+
• Flower (pip install flwr)
• Required ML/DL libraries (e.g., TensorFlow, PyTorch, NumPy, etc.)
• MIT-BIH Arrhythmia dataset

Installation

1. Clone the repository:

   git clone <your-repo-url>
   cd Arrythmia-Detection

2. Install dependencies:

   pip install flwr
   # Add other dependencies as needed

Usage

1. Prepare Data

Run prep_mitbih.py to preprocess the MIT-BIH dataset for clients.

2. Start the Server

python server.py

• The server listens on 0.0.0.0:8000 by default.

3. Start Clients

In separate terminals, start each client:

python client1.py
python client2.py

• Each client loads its data, trains locally, and communicates with the server.

Customization

• Modify client_base.py to change client logic or add new clients.
• Adjust server parameters (e.g., number of rounds) in server.py.
• Update data preprocessing steps in prep_mitbih.py as needed.

Monitoring & Logging

• The server and clients use Python's logging module for event tracking.
• Consider integrating external monitoring tools for production deployments.

Security

• For production, enable SSL/TLS for secure communication.
• Implement authentication for client-server connections if needed.

References

• Flower Documentation
• MIT-BIH Arrhythmia Database

License

Specify your license here (e.g., MIT, Apache 2.0).