However, actual problems usually have both linear and nonlinear
characteristics, and previous research attempted to explain
the variability of actual problems by combining two methods [27e
37]. Koutroumanidis et al. [35] predicted fuel wood prices in Greece
through a hybrid model that combined the ARIMA and the ANN
models. Tseng et al. [36] predicted the total production of the
machinery industry and the soft drink industry in Taiwan through a
hybrid forecasting model, known as SARIMABP model that combined
the SARIMA model and neural network BP (back propagation)
model. Shafie-khah et al. [37] predicted energy price using
ARIMA, radial basis function neural networks, and partial swarm
optimization.