|
Abstract:
|
Identifying change points and anomalies in dynamic network structures has become increasingly popular across various disciplines from cyber-security to blockchain to neuroscience. Most current statistical methods for anomaly detection on complex networks assume that network snapshots at different time points are independent. To address this limitation, we propose a new distribution-free framework for anomaly detection in dynamic networks. First, we present each network snapshot of the data as a linear object and find its respective univariate characterization (e.g. mean degree or clustering coefficient). Second, we adopt a change point detection method for (weakly) dependent time series based on efficient scores and enhance its finite sample properties by approximating the asymptotic distribution of the test statistic with sieve bootstrap. We apply our method to simulated data and to two fMRI datasets. Our studies suggest that the new approach delivers more accurate and interpretable results in terms of enhancing our understanding of dynamic brain region interactions than the competing techniques.
|
