Detecting fraud in financial transactions requires models that capture both behavioral anomalies and structural irregularities. This project combines gradient boosting methods with graph neural networks (GNNs) to improve detection accuracy. Gradient boosting (e.g., XGBoost, LightGBM) handles engineered tabular features, while GNNs uncover hidden patterns in transaction networks by modeling users and accounts as nodes. Together, these methods address class imbalance, reduce false positives, and detect collusive structures such as money-laundering rings. The goal is to build a scalable, interpretable fraud detection system capable of securing millions of transactions daily.

I joined SuperUROP because I want to see how the algorithms I’ve studied in class can make a real impact on financial security. This project lets me treat fraud as a hidden pattern to uncover, and I’m most excited to learn how graph neural networks can reveal connections that people alone might miss.”