Weighted recurrence networks for the analysis of time-series data

Recurrence networks (RNs) have become very popular tools for the nonlinear analysis of time-series data. They are unweighted and undirected complex networks constructed with specific criteria from time series. In this work, we propose a method to construct a 'weighted recurrence network' from a time series and show that it can reveal useful information regarding the structure of a chaotic attractor which the usual unweighted RN cannot provide. Especially, a network measure, the node strength distribution, from every chaotic attractor follows a power law (with exponential cut off at the tail) with an index characteristic to the fractal structure of the attractor. This provides a new class among complex networks to which networks from all standard chaotic attractors are found to belong. Two other prominent network measures, clustering coefficient and characteristic path length, are generalized and their utility in discriminating chaotic dynamics from noise is highlighted. As an application of the proposed measure, we present an analysis of variable star light curves whose behaviour has been reported to be strange non-chaotic in a recent study. Our numerical results indicate that the weighted recurrence network and the associated measures can become potentially important tools for the analysis of short and noisy time series from the real world.