Model-Based Anomaly Detection in Networks

Anomaly detection in network data is an important problem with applications ranging from fraud and cybersecurity to social and biological networks. The project aims to develop a statistically principled approach to anomaly detection based on probabilistic models for network structure. The main idea is to exploit model-based community detection methods, such as Stochastic Block Models (SBMs), to characterize the expected connectivity patterns of a network. Once the latent community structure and model parameters have been estimated, anomalous nodes or edges can be identified by measuring how strongly their observed connections deviate from those predicted by the fitted model. The resulting framework will provide interpretable anomaly scores grounded in an explicit statistical model of network structure.