Persistent Homology of Time Series through Complex Networks
A unified pipeline for univariate time series classification via complex networks and persistent homology is presented, showing that no single construction dominates and persistence-based features degrade gracefully under noise, consistent with the classical stability theorem of persistent homology.