Skip to content
Federated Learning for Anomaly Detection
Jan 2025 - May 2025

Federated Learning for Anomaly Detection

Privacy-preserving fraud detection with 91% accuracy

01. The Challenge

Financial institutions need fraud detection without sharing sensitive customer data across organizations.

02. The Solution

Implemented federated learning system enabling collaborative model training while preserving data privacy.


Overview

In the financial sector, data privacy is paramount. This project demonstrates how we can train powerful fraud detection models across multiple institutions without ever sharing the raw transaction data.


Methodology

Using FedML, we established a central server that coordinates the training process. Each participating node (bank) trains a local model on its private data and sends only the model updates (gradients) to the central server.


Privacy Mechanisms

  • **Differential Privacy:** Added noise to gradients to prevent reverse-engineering of data.
  • **Secure Aggregation:** The central server only sees the aggregated updates, not individual contributions.

    • Results

    The system achieved 91% accuracy in detecting fraudulent transactions, comparable to centralized training but with zero data leakage.


    System Architecture

    FedML framework for distributed training
    gRPC for secure communication
    TensorFlow for model architecture
    Differential privacy mechanisms

    Project Links

    Technologies

    FedMLTensorFlowgRPCNumPyPython

    Key Impact

    91% fraud detection accuracy
    Zero data sharing required
    Privacy-preserving ML