Robustness of the recurrence network analysis method with respect to data loss
Complex network theory applied to various dynamical systems has opened for quantitative analysis of the underlying structure of real-world phenomena. One useful method called the recurrence network analysis maps point processes based from their spatial, temporal and magnitude information into a network without much a priori assumptions about the system. Network statistics uncovers spatio-temporal dynamics providing insight into the causality mechanism of the system. However, this method is dependent on records of empirical data and thus data completeness remains an unaddressed issue. In this work, we apply recurrence network analysis into 3 different synthetic data sets with known spatio-temporal dynamics and use spectral graph analysis to characterize the properties of the network method. Results show network eigenvalue robustness up to 90% loss suggesting preservation of network properties despite data incompleteness for the network following a nearest spatial criteria. Moreover, it was found that the more complex the organization of events of a process in space and time, the closer the system’s recurrence network structure is to that of recurrence network for randomly occurring events.