Change point detection in complex networks based on weighted graphs

Complex networks are powerful tools for characterizing the temporal evolution of complex systems and are widely applied in fields such as finance and transportation. Structural changes frequently occur in such networks, making statistical change point detection essential for revealing and understanding these transitions. Existing graph-based methods, however, often rely primarily on edge counts, limiting their ability to capture structural changes comprehensively. To address this limitation, this paper proposes a weighted-graph-based method for change point detection in complex networks. The method highlights the importance of edge weights and constructs a novel network measure, termed the edge-weight similarity statistic, to enable single change point detection. Furthermore, a two-stage method is developed for multiple change point detection: first, the weighted-graph-based seeded binary segmentation(WG-SBS) algorithm is introduced to identify candidate change points; second, a new change point scoring function is defined and combined with a backward elimination(BE) strategy to select the optimal set of change points. A dendrogram is employed to enhance visualization and improve our understanding of the hierarchical structure of change points. Simulation studies and real data analysis demonstrate that the proposed method can effectively detect change points in networks.