Machine learning techniques to identify synchronization patterns in multiple timescale dynamical systems networks

• Synchronization patterns distribution of network models of coupled dynamical systems. • Combination of machine learning techniques with dynamical analysis. • Link between model behaviors and patterns in experimental data. We present a novel methodology that combines machine learning techniques with dynamical analysis to classify and interpret the behavior distribution of network models of coupled dynamical systems. Our methodology determines the optimal number of distinct behaviors and classifies them based on time-series features, allowing for an interpretable and automated partition of the parameter space. Applying this approach to a homogeneous two-clusters model of intracellular calcium concentration dynamics, we identify nine different long-term behaviors, including complex and chaotic regimes, mapping experimental data available in the literature. The results highlight the complementarity between data-driven classification and classical dynamical analysis in capturing rich synchronization patterns and detecting subtle transitions in multiple timescale biological systems.