The SARIMA model and the ANN model are methodologies that
predict future values using historically observed data, each of
which is suitable for linear and non-linear problems, respectively.
The SARIMA model cannot effectively explain non-linear problems,
while the ANN model offers sufficient explanatory power for nonlinear
problems [58]. On the other hand, the ANN model cannot
effectively explain the characteristics of seasonal and trend time
series [22]. As a result, if the SARIMA model or the ANN model is
implemented individually, it will be difficult to solve complicated
problems. Generally, however, observed values (Zt) have both a
linear factor (LFt) and a non-linear factor (NFt) (refer to equation
(9)) [37]. Therefore, it is necessary to develop a hybrid model that
combined both linear and non-linear approaches to solve this issue.