Robust Fraud Detection in E-Commerce with Explainable AI Techniques

Robust Fraud Detection in E-Commerce with Explainable AI Techniques

• Ruiqing Xu, Stuart Ph.D. student

Abstract:

In e-commerce, fraud detection is essential for trust and security. This research develops a robust fraud detection system using XGBoost, SVM, and Logistic Regression, enhanced with advanced Explainable AI (XAI) techniques—SHAP, LIME, Anchors, and LORE. These methods improve robustness by defending against noise, bias, and concept drift, while ensuring reliable predictions by minimizing variance in explanations over time.

The system adapts to evolving fraud patterns through incremental learning, providing dynamic updates. SHAP and LIME deliver local and global interpretability, Anchors generate precise rule-based explanations, and LORE offers counterfactual and surrogate rules. A comparative analysis evaluates the XAI techniques on metrics like interpretability, consistency, and computation time.

By integrating XAI, the system enhances transparency and resilience, offering a scalable solution for real-time e-commerce fraud detection with interpretable insights.

All Illinois Tech faculty, students, and staff are invited to attend.

The Friday Research Presentations series showcases ongoing academic research projects conducted by Stuart School of Business faculty and students, as well as guest presentations by Illinois Tech colleagues, business professionals, and faculty from other leading business schools.