Applying LSTM to time series predictable through time-window approaches

Abstract

Long Short-Term Memory (LSTM) is able to solve many time series tasks unsolvable by feed-forward networks using fixed size time windows. Here we find that LSTM's superiority does not carry over to certain simpler time series prediction tasks solvable by time window approaches: the Mackey-Glass series and the Santa Fe FIR laser emission series (Set A). This suggests to use LSTM only when simpler traditional approaches fail.