Publication Date

Spring 2026

Degree Type

Master's Project

Degree Name

Master of Science in Computer Science (MSCS)

Department

Computer Science

First Advisor

Fabio Di Troia

Second Advisor

Thomas Austin

Third Advisor

William B. Andreopoulos

Keywords

Machine learning, financial fraud detection, adversarial attacks, security, ML models

Abstract

Machine learning models are used in financial fraud detection to analyze large volumes of transaction related data and identify suspicious behavior. Techniques such as gradient-boosted decision trees and neural networks achieve strong performance on high-dimensional and highly imbalanced datasets. However these machine learning models are usually tested with the idea that people who commit fraud will not try to trick the system. Inthisprojectwestudytheadversarialrobustnessoffinancialfrauddetectionmod- els such as Logistic Regression, Random Forest, XGBoost, MLP, and FT-Transformer. Multiple classifiers are trained on the IEEE-CIS Fraud Detection dataset and subjected to ten adversarial attack types spanning gradient-based, optimization-based, univer- sal, boundary-based, and query-based methods. Our experiments show that model performance collapses significantly under these attacks. To address this, we apply TRADES-based adversarial training to recover robustness and develop a multi-class attack classifier capable of identifying the attack type. This study highlights the gap between clean accuracy and adversarial robustness in tabular financial fraud detection and shows that TRADES-style training can significantly improve robustness.

Download

Available for download on Saturday, May 22, 2027

Share

COinS