Credit Card Fraud Detection - Autoencoder

A comparative study of unsupervised anomaly detection techniques using PyTorch. This project evaluates different Autoencoder architectures and PCA to identify fraudulent transactions by measuring reconstruction error.

Authors

Project Overview

In the financial industry, fraudulent transactions are rare events (anomalies). This dataset contains transactions made by European cardholders, where frauds make up a small percentage of all transactions.

An Autoencoder is trained to compress and then reconstruct normal transactions with high precision.
When the model encounters a fraudulent transaction (which it hasn't seen before), it fails to reconstruct it accurately.
A high Mean Squared Error (MSE) between the input and the output indicates a high probability of fraud.

Model Architectures

Model	Architecture	Key Characteristic
Basic AE	30 → 15 → 7	Standard bottleneck design to learn essential features.
Deep AE	30 → 20 → 10 → 5	Deeper hierarchy with Dropout (0.2) to prevent overfitting.
Sparse AE	30 → 15 → 7	Uses L1 Regularization to force the model to use only the most important neurons.
PCA	Linear Baseline	Principal Component Analysis with 10 components for statistical comparison.

Tech Stack

Deep Learning: PyTorch
Data Manipulation: Pandas, NumPy
Machine Learning: Scikit-Learn (StandardScaler, PCA, Train-Test Split)
Visualization: Matplotlib
Dataset Handling: KaggleHub

Installation & Usage

Clone the repository:

git clone https://github.com/Jassiko6/credit-card-fraud-detection
cd credit-card-fraud-detection

Install dependencies:

pip install torch pandas numpy matplotlib scikit-learn kagglehub

Run the script:
```
python main.py
```

Downloads last month: -; Downloads are not tracked for this model. How to track

Inference Providers NEW

This model isn't deployed by any Inference Provider. 🙋 Ask for provider support