Credit Card Fraud Detection – Federated Learning

This project showcases the use of Federated Learning via TensorFlow Federated (TFF) to detect fraudulent credit card transactions while preserving data privacy. It also compares the federated approach with traditional centralized training using scikit-learn.

---

Project Overview

Credit card fraud detection requires analyzing transaction patterns to identify suspicious behavior. However, centralized machine learning models may compromise user privacy. Federated Learning (FL) enables decentralized model training across multiple clients, allowing raw data to remain local.

This project is organized into two main Google Colab notebooks:

Notebooks

1. Federated Learning Implementation

   • Built using TensorFlow Federated (TFF).
   • Simulates multiple clients training a model collaboratively without sharing their raw data.
   • Demonstrates secure and privacy-aware model training.

2. Centralized Credit Card Fraud Detection

   • Implements a baseline model using scikit-learn.
   • Provides data preprocessing, training, and evaluation for comparison with the federated model.

---

Dataset

• Dataset: Credit Card Fraud Detection Dataset (Kaggle)
• Features are anonymized for privacy.
• Highly imbalanced dataset with only ~0.17% fraudulent transactions.

---

Technologies Used

• Python
• TensorFlow Federated (TFF)
• TensorFlow
• scikit-learn
• Google Colab

---

Features

• End-to-end implementation of federated learning for fraud detection.
• Centralized vs. federated model performance comparison.
• Emphasis on data privacy and decentralized training.

---

Future Enhancements

• Integrate advanced models ANN in the federated setup.
• Apply class balancing techniques like SMOTE or weighted loss functions.
• Explore differential privacy for added security.