Grouping in Singular Spectrum Analysis of Time Series

Abstract

Singular spectrum analysis (SSA) is a nonparametric model-free time-series analysis and filtering technique with a wide variety of applications in numerous data-intensive fields. The grouping stage is the most crucial step in SSA, where the analyst selects significant components from the time series for further processing. However, there is no universal rule in this stage of grouping and the components need to be grouped based on the data characteristics. In this study, a few methods that can be adopted for grouping are discussed and their efficiencies in reconstructing the time series are compared. The results of the study will be helpful in understanding the procedure and will act as a guide in the selection of a method for grouping based on the data characteristics. Real-world daily rainfall time-series data were used as a case study.