Autoregressive models & deep learning(rnn-lstm) models both are used for time series prediction. As we choose the 'look back' for lstm's, provision to choose optimal lag by viewing acf-pacf plot or AR coefficient is also there for Autoregressive models.

  • When can we know that we have to use deep learning instead of other time series methods?
  • How do AR model & LSTM model differ? I know lstm is composed with gates & inherent memory blocks unlike AR, but theoretically why should not I use AR models?

On an other note, i would also like to know why RMSProp optimizer is used(or better) in recurrent neural networks?


AR models are linear and thus of limited flexibility when it comes to approximating nonlinear time series, but they are simple, easy to explain, and quick to estimate. LSTM are nonlinear and hence flexible, but they are complicated and computationally intensive. Therefore:

  • When you need fast estimation or ease of explanation, or have little data, or your phenomenon is well described as a linear function of past values, use AR.
  • When you do not care much about estimation speed and transparency, have lots of data, and the process is likely to be nonlinear, use LSTM.
| cite | improve this answer | |

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.