Page 182 - Contributed Paper Session (CPS) - Volume 6
P. 182
CPS1867 Winita S. et al.
Table 1: Comparison of RMSEs, MAEs, MAPEs, and MRAEs for the testing data
of accidental death obtained by hybrid SSA-NN and other methods
in literature
Based on Table 1 and Figure 3 (bottom) we can see that the hybrid SSA-NN
produce the smallest RMSE, MAE, MAPE, and MRAE among others.
Furthermore, Figure 3 (top) shows that the forecasting values for January 1979
until June 1979 obtained by SSA-NN red line) are the closest one to the actual
values (black line).
Figure 3: Comparison of actual and the forecast values obtained by several methods (top) and
comparison of the forecast accuracy between them (bottom).
4. Discussion and Conclusion
The combination between SSA and NN on monthly accidental deaths series
produces more accurate forecast values rather than Naïve, SARIMA, Subset
ARIMA, ARAR, and SSA-LRF. The SSA method decomposed the series into four
components, where the first three components can be approximated by the
deterministic functions. We model each component individually and then
combine them. In this step, we need to be careful in identifying and
determining the best fit function for the component series. By this technique,
171 | I S I W S C 2 0 1 9