Karunakar Thammadi

Fraud in financial systems is a very significant issue in the modern financial systems and it is becoming difficult to identify fraudulent transactions. Old ways of fraud detection usually look at the transactions one at a time and do not look at how accounts relate to each other. This project uses a network of financial transactions with nodes corresponding to accounts or transactions and edges corresponding to money transfers. Patterns of these transaction networks are learned using Graph Neural Networks in the form of a Directed Graph Convolutional Network (Directed GCN). The datasets used in the evaluation of the model included PaySim, Elliptic Bitcoin, and synthetic fraud dataset. Graph features were also added which helped in understanding the patterns of the transactions better. Precision, recall, F1 score and AUC are some of the metrics used to measure the performance of the model. Besides, a Streamlit dashboard was created in order to visualize the fraud predictions and transaction trends. These findings indicate that the graph models can be useful to model the relationships between transactions and assist in better fraud detection.

TBD