A Related Work

In this section, we will give an overview of the related literature in time series forecasting. Traditional Time Series Models The first generation of well-discussed time series model is the autoregressive family. ARIMA Box & Jenkins (1968); Box & Pierce (1970) follows the Markov process and build recursive sequential forecasting. However, a plain autoregressive process has difficulty in dealing non-stationary sequences. Thus, ARIMA employed a pre-process iteration by differencing, which transforms the series to stationary. Still, ARIMA and related models have the linear assumption in the autoregressive process, which limits their usage in complex forecasting tasks. Deep Neural Network in Forecasting With the bloom of deep neural networks, recurrent neural networks (RNNs) were designed for tasks involving sequential data.