Three-Model Deep Learning Fraud Detection System

A cutting-edge credit card fraud detection system leveraging three deep learning models: Multilayer Perceptron (MLP), Autoencoder (AE), and Variational Autoencoder (VAE). This system addresses the challenges of highly imbalanced datasets and evolving fraud patterns by combining supervised classification and unsupervised anomaly detection. Deployed via FastAPI for real-time predictions and includes an interactive dashboard.

⭐ Star this repository if you find it useful and share it with others!

Table of Contents

• Features
• Dataset
• Project Structure
• Installation
• API Endpoints
• Usage
• Technologies
• Contributing
• License
• Contact

Features

• Three Deep Learning Models:
  • MLP: Supervised classifier for direct fraud detection.
  • Autoencoder: Unsupervised anomaly detection based on reconstruction errors.
  • Variational Autoencoder: Probabilistic model for detecting subtle fraud patterns.
• Robust Preprocessing: Handles imbalanced data with SMOTE and scales features using RobustScaler.
• Real-Time Predictions: FastAPI-based REST API with endpoints for individual and ensemble model predictions.
• Interactive Dashboard: Web interface for testing fraud detection.
• Comprehensive Evaluation: Metrics include Accuracy, Precision, Recall, F1-score, ROC-AUC, and PR-AUC.

Dataset

• Source: Credit Card Fraud Detection (Kaggle)
• Features: 28 PCA-transformed anonymised features (V1–V28), transaction Amount, and Class (0 = non-fraudulent, 1 = fraudulent).
• Imbalance: Fraudulent transactions ≈ 0.17%.

Project Structure

Three-Model-Deep-Learning-Fraud-Detection-System/
├── app.py                    # FastAPI application
├── requirements.txt          # Project dependencies
├── README.md                 # Project documentation
├── data/                     # Raw dataset (ignored in git)
├── models/                   # Trained model files (ignored)
├── saved_models/             # Model checkpoints & scaler (ignored)
├── saved_artifacts/          # Preprocessed data & thresholds (ignored)
├── figures/                  # Plots and visualizations (ignored)
├── logs/                     # Training & API logs (ignored)
├── notebooks/                # Jupyter notebooks for EDA and experiments
├── tests/                    # Unit and integration tests
└── tuner_dir/                # Hyperparameter tuning artifacts

Installation

Prerequisites

• Python 3.8+
• Git
• Virtual environment (recommended)

Steps

1. Clone the Repository:

   git clone https://github.com/nafisalawalidris/Three-Model-Deep-Learning-Fraud-Detection-System.git
   cd Three-Model-Deep-Learning-Fraud-Detection-System

2. Create & Activate Virtual Environment:

   python -m venv fraud-dl-env

   • On Windows (PowerShell):

     fraud-dl-env\Scripts\Activate.ps1

   • On macOS/Linux:

     source fraud-dl-env/bin/activate

3. Install Dependencies:

   pip install -r requirements.txt

4. Run the FastAPI Application:

   uvicorn app:app --reload

5. Access the API:

   • Swagger UI: http://127.0.0.1:8000/docs
   • Dashboard: http://127.0.0.1:8000/dashboard

API Endpoints

Endpoint	Method	Description
/	GET	Root welcome message
/health	GET	API health check
/models	GET	List all available models
/predict/mlp	POST	Fraud detection using MLP
/predict/autoencoder	POST	Fraud detection using Autoencoder
/predict/vae	POST	Fraud detection using VAE
/predict/ensemble	POST	Fraud detection using ensemble voting
/dashboard	GET	Interactive dashboard for predictions

Usage

1. Preprocessing: The system handles duplicates, missing values, and scales data using RobustScaler. SMOTE is applied to address class imbalance.
2. Training: Models are implemented in PyTorch with dropout, batch normalisation, and early stopping.
3. Evaluation: Models are evaluated using Accuracy, Precision, Recall, F1-score, ROC-AUC, and PR-AUC.
4. Testing: Use the Swagger UI or cURL to send sample transactions for real-time predictions.

Technologies

• Language: Python
• Libraries:
  • Data: pandas, numpy
  • Visualisation: matplotlib, seaborn
  • Deep Learning: torch (PyTorch)
  • Model Persistence: joblib, pickle
• Deployment: FastAPI
• Version Control: Git + GitHub

Contributing

Contributions are welcome! Follow these steps:

1. Fork the repository.
2. Create a feature branch:

   git checkout -b feature-name

3. Commit your changes:

   git commit -m 'Add feature'

4. Push to your branch:

   git push origin feature-name

5. Open a Pull Request.

License

This project is licensed under the MIT License. See the LICENSE file for details.

Contact

• Portfolio: https://nafisalawalidris.github.io/13/
• GitHub: nafisalawalidris
• Email: Reach out for inquiries or collaboration.