[R] Confidence Intervals with mlp forecasts

[Nikolaos Kourentzes]

Hi Paul,
Currently it does not provide prediction intervals, as it is not assuming a generative model or a particular error distribution.
I think the best way forward, with nnfor, is to construct empirical ones. 
Have a look at this paper for some relatively straightforward approaches that work quite well under a variety of conditions. 
The paper looks at safety stocks, but the same approach can be used for generating prediction intervals.https://kourentzes.com/forecasting/2018/06/20/empirical-safety-stock-estimation-based-on-kernel-and-garch-models/
Best,Nikos


Professor, Dept. Management Science


Lancaster University Management School, UK
 
Centre for Marketing Analytics & Forecasting


blog: http://nikolaos.kourentzes.com
 
twitter: @nkourentz

Book: Ord, Fildes & Kourentzes (2017) Principles of Business Forecasting (2nd ed.), Wessex.
 

    On Tuesday, August 27, 2019, 3:44:45 PM GMT+3, Paul Bernal <paulbernal07 using gmail.com> wrote:  
 
 Dear friends,

Hope you are all doing well. I am currently using function mlp (to fit multiple layer percentron model) to generate forecasts using package nnfor.
I would like to know if the mlp function provides, or is there a way to construct confidence intervals for the forecasts generated by this mlp function.
Any help and/or  guidance would be much appreciated,
Best regards,
Paul  
	[[alternative HTML version deleted]]