Mazen Asag

Card Fraud Detection System 🕵️‍♂️💳

This project implements a robust end-to-end machine learning pipeline for detecting fraudulent credit card transactions.

🔍 Overview

• ⚙️ Modular pipeline for preprocessing, training, evaluation
• 🗂️ YAML-based configuration for cleaner code and experiments
• 📦 DVC for versioning datasets and pipeline stages
• 🚀 GitHub Actions for CI/CD automation

🗃️ Dataset Overview

Dataset: Credit Card Fraud Detection on Kaggle
Transactions: 284,807
Fraudulent: 492 (0.172%)

Features are anonymized using PCA, except 'Time', 'Amount', and 'Class' (target). Due to the imbalance, AUPRC is prioritized over accuracy.

🔁 Handling Class Imbalance

Used SMOTE (only on training set):

• Original: 227,845 × 31
• Post-SMOTE: 442,012 × 30

📁 Project Structure

CardFraud/
├── config/             # YAML configs
├── src/CardFraud/      # All modules
├── research/           # EDA Notebooks
└── .github/workflows/  # CI/CD
        

📦 DVC Versioning

• Version control for large files
• Tracked training outputs & reproducibility

⚙️ CI/CD Pipeline

Uses GitHub Actions to trigger ML workflows on push/merge events.

📊 Model Results Summary

Model	Accuracy	Precision	Recall	F1 Score
Logistic Regression	0.9803	0.0742	0.9072	0.1371
Random Forest	0.9995	0.8804	0.8350	0.8571
Decision Tree	0.9975	0.3878	0.7835	0.5187
XGBoost	0.9995	0.8586	0.8763	0.8673

🏆 Best Model: XGBoost

Best performing with highest F1-score.

🚀 How to Run

git clone https://github.com/Mazenasag/End-to-end-Detecting-Card-Fraud
cd CardFraud
pip install -r requirements.txt
python src/CardFraud/pipeline/main.py
        

✍️ Author

Developed by Mazen Asag