Time series outlier detection using the trajectory matrix in singular spectrum analysis with outlier maps and ROBPCA

Abstract

Singular spectrum analysis is a powerful non-parametric time series method that applies singular value decomposition to a Hankel structured matrix. The method can handle complex time series structures that include combinations of polynomials, sinusoids and exponentials. Outlier maps combined with robust principal component analysis is considered and shown to compare very favourably with existing time series methods to identify an additive time series outlier. The wellknown airline time series as well as a South African tourism time series are used to illustrate the usefulness of the methodology