Aegis: Fraud Detection with Deep Learning

Overview

Aegis is a financial fraud detection system built using deep learning.
The goal of this project is to identify fraudulent transactions based on transaction details, payment card data, device information, and engineered features.

By leveraging real-world data and modern AI techniques, Aegis aims to showcase how deep learning can help financial institutions proactively flag suspicious activity and reduce fraud losses.

---

Dataset

We use the IEEE-CIS Fraud Detection dataset from Kaggle.
This dataset contains anonymized transaction and identity information collected by Vesta’s fraud protection system.

Key features include:

• Transaction details (amount, product type, transaction delta time, etc.)
• Card data (issuer bank, card type, card category)
• Location and email info (address, purchaser/recipient email domains)
• Engineered features (counting, distances, entity relations, etc.)

---

Approach

1. Data Cleaning & Preprocessing

   • Rename cryptic columns for clarity.
   • Handle missing values and categorical encoding.
   • Normalize numerical features.

2. Model Training

   • Train a deep learning neural network to predict whether a transaction is fraudulent (isFraud).
   • Evaluate model performance with metrics like AUC, precision, recall, and F1-score.

3. Deployment (future work)

   • Integrate the trained model into a backend service (e.g., REST API).
   • Enable real-time fraud detection for streaming transactions.

---

Authors

• Shaivil Patel
• Ethan Espineli

---