Recurrence-based time series analysis by means of complex network methods

Complex networks are an important paradigm of modern complex systems sciences which allows quantitatively assessing the structural properties of systems composed of different interacting entities. During the last years, intensive efforts have been spent on applying network-based concepts also for the analysis of dynamically relevant higher-order statistical properties of time series. Notably, many corresponding approaches are closely related with the concept of recurrence in phase space. In this paper, we review recent methodological advances in time series analysis based on complex networks, with a special emphasis on methods founded on recurrence plots. The potentials and limitations of the individual methods are discussed and illustrated for paradigmatic examples of dynamical systems as well as for real-world time series. Complex network measures are shown to provide information about structural features of dynamical systems that are complementary to those characterized by other methods of time series analysis and, hence, substantially enrich the knowledge gathered from other existing (linear as well as nonlinear) approaches.

💡 Research Summary

The paper provides a comprehensive review of recent methodological advances that combine complex‑network theory with recurrence‑based analysis for time‑series data. Starting from the classical concept of a recurrence plot (RP) – a binary matrix that marks when two states in a reconstructed phase space are within a predefined distance ε – the authors reinterpret the RP as an adjacency matrix and thus construct a “recurrence network” (RN). In this network each node corresponds to a time index, while an edge represents a recurrence (i.e., two points are ε‑neighbors). By varying ε, the network can be made sparse or dense, and weighted versions can be built using the actual distances between points.

Once the RN is built, a full toolbox of graph‑theoretic measures is applied: average path length (L), clustering coefficient (C), global efficiency (E), degree distribution P(k), and modularity (Q). Each of these quantities captures a distinct aspect of the underlying dynamics. L and E are sensitive to the overall mixing properties of the system and tend to shrink when the dynamics become more chaotic. C reflects the prevalence of locally triangular structures in phase space, which is high for periodic or quasi‑periodic motion. P(k) can reveal scale‑free behavior, indicating self‑organization, while Q quantifies the presence of distinct dynamical modules (e.g., coexisting attractors). The authors compare these network metrics with traditional recurrence quantification analysis (RQA) statistics such as recurrence rate, determinism, and entropy, showing that the network approach provides complementary structural information that is not captured by scalar RQA measures.

The methodological discussion is illustrated with two families of examples. First, synthetic chaotic systems – the Lorenz attractor and the Hénon map – are examined. By scanning ε, the authors demonstrate that C and Q exhibit sharp increases at the transition from chaotic to periodic regimes, whereas L and E decrease, mirroring the change in the geometry of the attractor. These transitions are also visible in RQA’s determinism, but the network view makes the underlying topological re‑wiring explicit. Second, real‑world data are analyzed. For a climatological temperature series, seasonal cycles appear as separate modules; modularity peaks at seasonal boundaries, and clustering rises during periods of strong periodicity. For electrocardiogram (ECG) recordings, the onset of atrial fibrillation is preceded by a marked drop in clustering and a rise in average path length, suggesting that RN‑based metrics could serve as early warning indicators.

The authors are candid about the limitations of the approach. The choice of ε is somewhat arbitrary and can dramatically affect network topology; short time series lead to sparse networks with unreliable statistics; high‑dimensional embeddings increase computational cost to O(N²). To mitigate these issues they propose adaptive ε schemes (e.g., k‑nearest‑neighbor thresholds), multi‑scale network constructions, and parallel GPU implementations. They also suggest hybridizing RN analysis with other nonlinear tools such as transition matrices or permutation entropy, and exploring automatic parameter optimization.

In conclusion, recurrence‑based complex networks enrich time‑series analysis by providing a structural lens that complements conventional linear and nonlinear techniques. Network measures capture dynamical transitions, multi‑scale organization, and abnormal events in a way that is both quantitative and visually interpretable. The review highlights the broad applicability of the methodology across physics, biology, climate science, and engineering, and points toward future research directions including real‑time network updating, integration with deep learning, and systematic benchmarking against established statistical methods.