This question already has an answer here:

I'm doing Long-Short-Term-Memory (LSTM) to forecast time series. I was wondering, could we add x-reg part to an LSTM model? Like adding an X-Reg part to and ARIMA model?

Say I have a response monthly time series y (for about 10~20 years observation). And say I have about 400 signals (predictor time series). Should I do feature-selection and regression first and then use LSTM to model the residuals?

Or is this the wrong way to add predictors to LSTM?

Thank you!


marked as duplicate by Sycorax, kjetil b halvorsen, mdewey, Michael Chernick, jbowman Aug 1 '18 at 20:27

This question has been asked before and already has an answer. If those answers do not fully address your question, please ask a new question.

  • $\begingroup$ You can just add them as additional inputs to your LSTM. It's actually much more straightforward than with ARIMA or ETS. $\endgroup$ – Skander H. Jul 20 '18 at 21:54
  • $\begingroup$ Hi Alex, Could you give an example on how to add xreg to LSTM? if my "dim(y_train_arr)" is [105,1], and I have 2 predictors, what should my "dim(x_train_arr)" be? [105,3,1] or [105,1,3]? $\endgroup$ – Chenying Gao Jul 27 '18 at 14:37
  • $\begingroup$ That will depend on what too you are using for LSTM. $\endgroup$ – Skander H. Jul 30 '18 at 23:20