Graph Contrastive Learning for Fraud Detection in Financial Transactions Using AI-Powered Anomaly Detection in Banking and E-Commerce
Abstract
The rapid evolution of financial fraud in banking and e-commerce renders conventional detection techniques insufficient. This work presents a Graph Contrastive Learning (GCL) framework, FraudGraph-CL, meant to improve fraud detection in financial transactions. The suggested method demonstrates complex interactions among consumers, transactions, stores, and gadgets by simulating financial operations as a heterogeneous graph. By using contrasting learning, depictions of nodes are enhanced, thereby enabling strong fraud classification even in low-data or hostile settings. Utilizing graph augmentation techniques such edge tampering and feature masking to improve anomaly identification, FraudGraph-CL combines Graph Neural Networks (GNNs) with contrastive loss optimization. By 12.3% compared to state-of-the-art models, FraudGraph-CL increases fraud detection accuracy by 95.2%, 94.8% precision, and 96.1% AUC-ROC experimental outcomes on real-world financial datasets. Providing a scalable, real-time fraud protection system, the suggested framework is relevant in banking, digital payments, e-commerce platforms, and financial regulatory systems. FraudGraph-CL much improves fraud detection efficiency, interpretability, and adaptability against changing fraud techniques by using graph-based relational learning and self-supervised neural networks