Fraud Detection on Credit Card Transactions by hussein hafezFraud Detection on Credit Card Transactions by hussein hafez

Fraud Detection on Credit Card Transactions

hussein hafez

Completed work

ML Engineer

Matplotlib

pandas

scikit-learn

Investment & Asset Management

Fraud Detection on Credit Card Transactions

A machine learning pipeline to detect fraudulent credit card transactions using imbalanced classification techniques.

Built with scikit-learn, this project demonstrates a complete workflow from preprocessing to model tuning and evaluation — focused on real-world constraints like class imbalance, interpretability, and performance metrics.

Problem Overview

Goal: Classify transactions as fraudulent or legitimate.

Challenge: Only ~0.17% of transactions are fraud — a highly imbalanced binary classification problem.

Dataset: Kaggle – Credit Card Fraud Detection

Technologies & Libraries

Python 3.9

scikit-learn

imbalanced-learn (SMOTE)

matplotlib / seaborn

pandas / numpy

ML Pipeline Summary

Data Preprocessing

Standardized features using StandardScaler

Stratified train/test split to preserve class imbalance

Handling Class Imbalance

Applied SMOTE (Synthetic Minority Over-sampling) to balance the dataset before training

Model & Training

Used Random Forest as the primary classifier

Performed hyperparameter tuning with GridSearchCV (5-fold, ROC-AUC scoring)

Evaluation

ROC-AUC Score: Measured on test set

Precision-Recall Curve: Emphasizes performance on rare fraud cases

Confusion Matrix: Visualizes false positives/negatives

Classification Report: Full precision, recall, F1 breakdown

Feature Importance

Visualized key fraud indicators using feature_importances_ from the trained model

Results

| Metric | Value |

|---------------|-----------|

| ROC-AUC Score | ~0.97 |

| PR AUC | High separation between classes

| Top Features | V14, V10, V17, V12

What This Project Demonstrates

Ability to build real-world ML pipelines end-to-end

Knowledge of class imbalance handling using SMOTE

Comfort with evaluation beyond accuracy (PR, ROC, Confusion Matrix)

Awareness of model interpretability with feature importance

How to Run

Clone the repo

git clone https://github.com/husseinhafez1/Fraud-Detection-on-Credit-Card-Transactions
cd fraud-detection-ml

Install dependencies

pip install -r requirements.txt

Launch notebook

jupyter notebook

Run fraud_detection_pipeline.ipynb

File Structure


fraud-detection-ml/
├── data/
│   └── creditcard.csv                   # Dataset (Kaggle)
│
├── notebooks/
│   └── fraud_detection_pipeline.ipynb   # Main notebook
│
├── src/
│   └── preprocessing.py                 # (Optional) data processing logic
│
├── fraud_detection_pipeline.html        # HTML export of notebook
├── requirements.txt                     # Dependencies
└── README.md                            # Project description

Like this project

Completed work

Posted Aug 16, 2025

Developed a fraud detection ML pipeline for credit card transactions.

Likes

Views