I am analyzing time series data on electricity prices using a variety of methods (ARIMAX, TBATS, STLM) including the auto-regressive feed-forward neural network NNETAR, which is implemented in the forecast package.

Since auto.arima, tbats etc. choose parameters automatically, I've let nnetar choose the parameters automatically respectively I used the default values as well.

I only have a training set and a test set, but I have read, that it can be beneficial to further split the training set into a smaller training set and a validation set. The model is then tested on the validation set. After the optimal parameters have been obtained, a model is fitted on the "whole" training set (small training set + validation set) and tested on the test set.

Is there an implementation for auto-regressive time series parameter tuning in R (many times parameter tuning involves the caret package)?

  • $\begingroup$ You'd better try the keras implementation for R. Nnetar gives subpar performance in time series forecasting - it lacks layers, you cannot choose activation functions, learning rate, etc. Since nnetar comes with so few parameters - why not test them in the loop? $\endgroup$ – SWIM S. Jul 19 at 16:51

Your Answer

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

Browse other questions tagged or ask your own question.