Credit card fraud detection remains one of the most pressing challenges in financial technology, costing institutions an estimated 32 billion annually. This paper proposes GNN-XGB, a hybrid framework combining Graph Neural Networks with XGBoost to exploit individual transaction features andgraph-neighborhood patterns. We construct a transaction graph of 10, 000 nodes and 629, 820 edges, apply two-layer mean-aggregation message passing, and feed graph features into XGBoost. Our framework achieves F1 Score of 96. 99% compared to 49. 15% baseline, a +47. 84% improvement. Graph featuresaccount for over 90% of predictive power, confirming fraud is a network phenomenon. Implementation uses pure Python with no specialized libraries.
vikram Paritosh (Thu,) studied this question.