Fraud Detection In Financial Transaction Using GNN With XGBoost

Presently fraud detection mechanisms fast, accurate are vastly more critical to financial technology, online transactions, where speedy action through very early detection of fraudulent transactions and acts can safeguard assets and prevent tremendous amounts of losses. This paper puts forward a new hybrid model using GNNs in combination with XGBoost to enhance precision in fraud detection and capture complex patterns as well as relational dependencies involved in financial networks. The GNN component makes use of the graph structure to capture complex interactions among entities based on frequency, location, and time features representing transaction relationships. It subsequently uses XGBoost to fine tune and boost the accuracy for classification purposes based on its robustness and scalability. Combined into the above methods, our model allows for a subtle analysis of transaction behaviors and is able to differentiate between legitimate and fraudulent transactions with a higher degree of precision than the traditional approaches. An experimental application on benchmark datasets gives evidence that our hybrid model becomes more effective in improving fraud detection rates with a minimum number of false positives and, thus, constitutes an appropriate element for realizing real-time fraud prevention systems.