Overview
Led by Caterina De Bacco, Hadiseh Safdari, and Martina Contisciani, this project investigated how to detect unusual patterns in complex networks. The project was supported by the Cyber Valley Research Fund from September 2021 until December 2023.
The researchers developed probabilistic models that combine anomaly detection with community detection. In this context, communities are groups of nodes—such as users, organizations, or transport hubs—that interact more frequently with one another. Interactions that do not fit these regular patterns can then be identified as potential anomalies.
The project examined static, reciprocal, and dynamic networks, in which connections appear and disappear over time.
Key results
- Developed models that infer both hidden community structures and the probability that an edge is anomalous.
- Showed that anomalies can be detected using network structure alone, without requiring additional information about individual nodes.
- Extended the model to account for reciprocity, meaning that connections in opposite directions—such as mutual communication—can influence what counts as regular behavior.
- Developed DynACD, a model for dynamic networks that uses temporal information to identify unusual changes in interactions.
- Demonstrated that the dynamic model performed better than several comparison methods in detecting anomalous edges, particularly when the timing of interactions was important.
- Showed that identifying and removing anomalous edges can improve community detection and that the model can also identify unexpected or potentially missing connections.
- Released open-source software implementations, including ACD and CRAD.
This project has yielded three peer reviewed publications:
- “Anomaly detection and community detection in networks”, H. Safdari and C. De Bacco, J Big Data 9, 122, https://doi.org/10.1186/s40537-022-00669-1, 2022.
- “Anomaly, reciprocity, and community detection in networks”, H. Safdari, M. Contisciani and C. De Bacco, Phys. Rev. Research 5, 033084, https://link.aps.org/doi/10.1103/PhysRevResearch.5.033084, 2023.
- “Community detection and anomaly prediction in dynamic networks”, H. Safdari and C. De Bacco, Communications Physics, 7, 397. doi:10.1038/s42005-024-01889-y, 2024.
