Recurrent Neural Networks for Time Series Forecasting

Recurrent neural networks are well suited to supervised learning problems where the dataset has a sequential nature. Time series forecasting should not be an exception. RNNs are essentially neural networks with memory. They can remember things from the past, which is obviously useful for predicting time-dependent targets. Yet, applying them to time series forecasting is not a trivial task. The aim of this article is to review some practices that can help RNNs deliver useful forecasts.