The Nonlinear autoregressive exogenous (NARX) model, which predicts thecurrent value of a time series based upon its previous values as well as thecurrent and past values of multiple driving (exogenous) series, has beenstudied for decades. Despite the fact that various NARX models have beendeveloped, few of them can capture the long-term temporal dependenciesappropriately and select the relevant driving series to make predictions. Inthis paper, we propose a dual-stage attention-based recurrent neural network(DA-RNN) to address these two issues. In the first stage, we introduce an inputattention mechanism to adaptively extract relevant driving series (a.k.a.,input features) at each time step by referring to the previous encoder hiddenstate. In the second stage, we use a temporal attention mechanism to selectrelevant encoder hidden states across all time steps. With this dual-stageattention scheme, our model can not only make predictions effectively, but canalso be easily interpreted. Thorough empirical studies based upon the SML 2010dataset and the NASDAQ 100 Stock dataset demonstrate that the DA-RNN canoutperform state-of-the-art methods for time series prediction.
展开▼
