Think Globally, Act Locally: A Deep Neural Network Approach to High-Dimensional Time Series Forecasting

Forecasting high-dimensional time series plays a crucial role in many such as demand forecasting and financial predictions. Modern can have millions of correlated time-series that evolve together, i. e are extremely high dimensional (one dimension for each individual-series). There is a need for exploiting global patterns and coupling them local calibration for better prediction. However, most recent deep approaches in the literature are one-dimensional, i. e, even though are trained on the whole dataset, during prediction, the future forecast a single dimension mainly depends on past values from the same dimension. this paper, we seek to correct this deficiency and propose DeepGLO, a deep model which thinks globally and acts locally. In particular, is a hybrid model that combines a global matrix factorization model by a temporal convolution network, along with another temporal that can capture local properties of each time-series and associated. Our model can be trained effectively on high-dimensional but time series, where different time series can have vastly different, without a priori normalization or rescaling. Empirical results that DeepGLO can outperform state-of-the-art approaches; for, we see more than 25% improvement in WAPE over other methods on a dataset that contains more than 100K-dimensional time series.