Dynamic Spectral Clustering

Abstract

Clustering is a powerful tool for data classification; however, its application has been limited to analysis of static snapshots of data which may be time-evolving. This work presents a clustering algorithm that employs a fixed time interval and a time-aggregated similarity measure to determine classification. The fixed time interval and a weighting parameter are tuned to the system’s dynamics; otherwise the algorithm proceeds automatically finding the optimal cluster number and appropriate clusters at each time point in the dataset. The viability and contribution of the method is shown through simulation