Abstract:
Enterprise Resource Planning (ERP) systems are essentially the backbone of modern
businesses, playing a crucial role in managing their daily operations. Yet, this complex and
high-volume nature of financial transactions observed within an ERP system on a daily basis
makes them a prime target for numerous types of fraud and anomalies. Effective detection and
subsequent transparent analysis are crucial in mitigating potential losses and operational
disruptions to business operations.
Within this paper, the author proposes a novel fraud detection framework leveraging a
Variational Autoencoder (VAE) working in tandem with an XGBoost (XGB) model for
anomaly identification, detection and enhanced interpretability. The VAE learns a compressed
representation of normal transaction patterns, enabling the identification of deviations
indicative of anomalies. Subsequently, XGBoost, a supervised learning algorithm, refines
these anomaly scores, providing a more accurate classification. Furthermore, the framework
incorporates Explainable AI (XAI) techniques to enhance transparency and interpretability.
