Enhancing Financial Data Traceability Using Graph Neural Networks for Provenance Methods

In this work, the application of Graph Neural Networks (GNNs) for improving financial data traceability is investigated in the aspects of accuracy, scalability, and realism. Financial transaction data was modeled as graphs, and the nodes were the entities of the financial transactions while the edges were the transactions themselves. The data source employed in the study involved 1,000,000 transactions made by 100,000 distinct entities from a large global financial organization. The approach that was used in this study involved a GNN model that incorporated the Graph Convolutional Network (GCN) structure, which was developed using the PyTorch Geometric library on a computer that was equipped with an NVIDIA Tesla V100 GPU. There are clear gains with respect to previous approaches, where the GNN model had an accuracy of 91%. 3%, precision of 89. 7%, recall of 90. Specifically, the precision is 95%, recall 85%, and F1-score of 90. 1% whereas, the accuracy of the current study using the logistic regression model was 78. CART showed a slightly better performance with 82. 1% accuracy while the random forests displayed 4% accuracy. The results for the second indicator, the traceability completeness, were improved from 75. 6% to 92. 4% and thus stressing on the practical advantages that can be derived from the approach. These results prove that the proposed GNN-based methodology is indeed effective, efficient, and scalable and may become a valuable tool for improving the financial data traceability and security. Future work will involve refining the current system, determining new uses for the system, and identifying new issues to address so that the system can be fine-tuned and expanded to meet the needs of the organization.