Wang, Cheng-HuaCheng-HuaWangWen, Tzai-HungTzai-HungWen2026-01-232026-01-232026-01https://www.scopus.com/pages/publications/105024306422https://scholars.lib.ntu.edu.tw/handle/123456789/735553Epidemic dynamics are effectively modeled using tree structures that represent the transmission pathways of infectious diseases. Analyzing these transmission trees offers key insights into the mechanisms driving power-law growth, especially in the early stages of an outbreak. However, the degree of self-similarity within these trees remains insufficiently understood. This study introduces a Bayesian Connectivity Algorithm (BCA) designed to detect Conditional Infectious Connectivity Structures (CICSs) that exhibit self-similar properties. Simulation results demonstrate that the BCA accurately identifies self-similar patterns, except in cases of high homogeneity in transmission abilities or shallow tree generations. Empirical analysis of Canadian COVID-19 data further confirms the robustness of these self-similar structures, even in the presence of disruptions in transmission chains, aligning with observed power-law growth. In summary, the BCA detects self-similarity in early epidemic transmission stages by identifying CICSs within transmission trees, providing critical insights for forecasting epidemic trajectories and characterizing early transmission dynamics.Bayesian inferenceComplex social systemsEarly stageEpidemiologyPower-lawSelf-similarity[SDGs]SDG2journal article10.1016/j.physa.2025.131179