| October 1, 2024
Project overview
The project, conducted within the Lusis Chair on AI and Finance (https://chaire-lusis.centralesupelec.fr/), focuses on credit card fraud detection, a growing challenge with global financial losses reaching $33.5 billion in 2022. The research aims to evaluate the applicability and performance of Stream Graphs [1, 2] for fraud detection, specifically through the lens of anomaly detection.
Unlike traditional fraud detection methods, this study explores dynamic graph-based representations to model fraud patterns. Given access to large-scale transaction datasets (600 million transactions), the proposed techniques must be highly scalable to handle real-world financial data efficiently.
Students will first familiarize themselves with Stream Graphs, an extension of classical graph structures incorporating temporal dynamics. The next step involves identifying fraud-relevant features derived from this dynamic graph representation and conducting a comparative analysis with existing statistical fraud detection methods. The project will then focus on developing classification and anomaly detection techniques based on these newly identified features and evaluating their effectiveness. A key objective is to determine whether dynamic graph modeling can help detect coordinated fraud campaigns.
Deliverables include:
- Modeling fraud data using Stream Graphs in Python/PyTorch.
- Visualization of fraud patterns and potential identification of fraud campaigns.
- Anomaly detection and classification models leveraging the Stream Graph representation.
- A final report documenting the modeling process, feature analysis, and performance evaluation of the proposed methods.
References
[1] Latapy, M., Viard, T., & Magnien, C. (2017). Stream Graphs and Link Streams for the Modeling of Interactions over Time (No. arXiv:1710.04073). arXiv. https://doi.org/10.48550/arXiv.1710.04073
[2] Latapy, M., Magnien, C., & Viard, T. (2019). Weighted, Bipartite, or Directed Stream Graphs for the Modeling of Temporal Networks (p. 49-64). https://doi.org/10.1007/978-3-030-23495-9_3
[3] Ranshous, S., Shen, S., Koutra, D., Harenberg, S., Faloutsos, C., & Samatova, N. F. (2015). Anomaly detection in dynamic networks: A survey. WIREs Computational Statistics, 7(3), 223-247. https://doi.org/10.1002/wics.1347