Simulation study of singular spectrum analysis from time series data with outlier

Abstract

In this simulation study, Singular Spectrum Analysis (SSA) will be used to analyze time series data with outlier. Time series data with outliers will be generated by open source software R (OSSR) for many replications. The goal of this research is to evaluate robustnes of the SSA model. SSA analyzes both types of time series data, with outlier and without outlier. Performance of S SA will be evaluated by changing of SSA parameter (Window length) and other parameter is determined by automatic grouping. Moreover, the MAPE (mean absolute percentage Error) of SSA is measured for both types of data. The result of Simulation study shows the larger the data the smaller the effect of the outliers. However, the larger the data the greater the shift in the value of L. The accuracy of SSA will decrease by 0.245 (MAPE) if there is available outlier in Moving Average (MA) data. For Autoregressive data, the accuracy of SSA will decrease by 0.264 (MAPE) if there is available single outlier in time series data.