Graph online change point detection based on fréchet statistics

Rui Luo, Jie Bao, Hing Cheung So, Suqun Cao, Mengqiao Xu

Abstract: Highlights•Propose an efficient online change point detection for dynamic graphs using Fréchet statistics of graph Laplacians.•Construct a metric space for graph Laplacians, addressing singularity and non-Euclidean nature of graph data.•Develop an incremental CUSUM-style test with theoretical guarantees on false alarm control and detection consistency.•Introduce novel incremental computation formulas, drastically reducing complexity and memory for large-scale applications.•Demonstrate superior detection performance and reduced false alarms in extensive numerical experiments and real-world applications.

External IDs:doi:10.1016/j.neucom.2025.131874