Likelihood-based anomaly detection in preferential attachment networks
Preferential attachment (PA) network is a widely used model for capturing the growth dynamics of real-world networks, in which newly arriving vertices are more likely to connect to existing vertices with higher degrees. In this paper, we consider a setting in which an anomalous vertex appears at some time point and receives edges with an additional attachment advantage governed by a parameter $\beta$, while ordinary vertices continue to follow the PA mechanism with parameter $\delta$. Detecting such anomalies is challenging due to the high variability in degree growth in PA networks and the limited information available when the anomaly arrives late in the network evolution. We propose an iterative parameter estimation procedure together with a likelihood-based detection framework. Simulation results show that the proposed procedure provides accurate estimation of the parameters $\beta$ and $\delta$. Detection performance depends on the time of anomaly occurrence: anomalies arising at midway stages of the network evolution are detected most reliably, whereas very early and late anomalies remain challenging, particularly when the parameter $\beta$ is small.