Page 177 - Contributed Paper Session (CPS) - Volume 6
P. 177
CPS1867 Winita S. et al.
On complex seasonal SSA based forecasting
1
1
1,2
3
Winita Sulandari ; Subanar ; Suhartono ; Herni Utami ; Muhammad Hisyam
4
Lee
1 Universitas Gadjah Mada, Yogyakarta, Indonesia
2 Universitas Sebelas Maret, Surakarta, Indonesia
3 Institut Teknologi Sepuluh Nopember, Surabaya, Indonesia
4 Universiti Teknologi Malaysia, Johor, Malaysia
Abstract
Modifications in the forecasting model based on SSA have been an interesting
subject among researchers in recent years. SSA is considered as a powerful
method in decomposing complex time series and a flexible model can be
constructed based on SSA decomposition. In this paper, a wellknown monthly
accidental deaths in USA is used as the experimental study. We compare the
performance of the hybrid SSA-model for the deaths series with other
methods proposed in the literature. The results show that the hybrid SSA-NN
model yields better forecasting accuracy than other methods discussed in the
literature.
Keywords
SSA; Time series; Accidental; Hybrid; NN
1. Introduction
The discussion of singular spectrum analysis (SSA) and its application has
been conducted by researchers from various fields (see Afshar & Bigdeli, 2011;
Hassani, Soofi, & Zhigljavsky, 2010; Hassani, Webster, Silva, & Heravi, 2015;
Mahmoudvand, Konstantinides, & Rodrigues, 2017; Nong, 2012; Suhartono et
al., 2018). This method can be helpful in extracting and identifying the trend,
harmonics and noise of a series (Hassani, 2007). SSA with linear recurrent
formula (LRF) discussed by Golyandina & Korobeynikov (2014) is more
appropriate for handling deterministic forecasting problem. Recently,
modelling each component of SSA by stochastic models and then combining
the results can be an alternative to SSA-LRF (Liu, Zhang, & Zhang, 2015).
By employing the hybrid method, the weakness of one method can be
handled by another method such that the accuracy performance of the
forecasting values usually improves. Neural network (NN) is a powerful
method in solving complex forecasting problems (G. P. Zhang & Qi, 2005).
Tseng, Yu, & Tzeng, (2002) has successfully combined NN with ARIMA and
applied this hybrid method to forecast seasonal time series data.
166 | I S I W S C 2 0 1 9