A Review of Machine Learning and Deep Learning Approaches for Fraud Detection Across Financial and Supply Chain Domains

Fraud detection has emerged as a critical challenge in modern digital economies, affecting both financial institutions and supply chain networks. The rapid proliferation of online transactions, coupled with increasingly sophisticated fraud schemes, has necessitated the development of advanced detection systems capable of identifying anomalous patterns in real-time. This systematic review examines the state-of-the-art machine learning and deep learning approaches for fraud detection across financial and supply chain domains, covering research published between 2015 and 2025. We analyze over 80 studies, categorizing methodologies into traditional machine learning, deep learning, ensemble techniques, semi-supervised learning, and emerging paradigms. Our review identifies key challenges including extreme class imbalance, concept drift, interpretability requirements, and real-time processing constraints. We evaluate the effectiveness of various algorithms using standard performance metrics and discuss their practical deployment considerations. The findings reveal that ensemble methods and tree-based models consistently achieve superior performance in credit card fraud detection, while semi-supervised approaches show promise for supply chain applications where labeled data is scarce. This review provides researchers and practitioners with a comprehensive understanding of current methodologies, their comparative strengths, and future research directions in fraud detection systems.