The HJ-biplot visualization of the Singular Spectrum Analysis method

Abstract

Time series data usually emerge in many scientific domains. The extraction of essential characteristics of this type of data is crucial to characterize the time series and produce, for example, forecasts. In this work, we take advantage of the trajectory matrix constructed in the Singular Spectrum Analysis, as well as of its decomposition through the Principal Component Analysis via Partial Least Squares, to implement a graphical display employing the Biplot method. In these graphs, one can visualize and identify patterns in time series from the simultaneous representation of both rows and columns of such decomposed matrices. The interpretation of various features of the proposed biplot is discussed from a real-world data set.