Analysing Trends and Forecasting of COVID-19 Pandemic in Malaysia using Singular Spectrum Analysis
The Singular Spectrum Analysis (SSA) is a powerful non-parametric time series analysis that has demonstrated its capability in forecasting different time series in various disciplines. SSA falls in the framework of data-driven modelling of dynamical system which does not rely on any underlying assumption except the inherent dynamics which are captured over time. The capabilities of SSA are mainly afforded by its direct connection to the singular value decomposition (SVD). It is generally accepted that SVD-based methods are very affective for the noise reduction in deterministic time series and consequently for forecasting, as well as for extracting trends and structures. Despite its strength, several shortcomings of SSA in the analysis of COVID-19 time series have been reported in the literature. The aim of this paper is to determine the scope of this limitation and we confine our investigation in the analysis and forecasting of COVID-19 Pandemic in Malaysia. We scrutinize the results fromthe SSA analysis of the number of daily confirmed cases to gain further insight into the intrinsic trends of the pandemic. Groupings of the singular spectra that contributes to different features of the pandemic time series are identified using analysis of the singular value spectrum, periodogram analysis and analysis of the weighted correlation matrix. It was revealed that under stationary conditions, the principal eigentriple is sufficient to produce reliable forecast. However, in non-stationary conditions, for example during a movement control order, it is useful to also study the minor eigentriples which could contain transient dynamics that may persist.