0
$\begingroup$

I once read about somebody who added noise to their time series before training a model. They didn't write why they did it though.

Is this common practice?

If it is, why do people do it ie. to prevent over-fitting?

$\endgroup$
4
  • $\begingroup$ Could you link the source? Thank you $\endgroup$
    – Leevo
    Oct 15, 2019 at 9:44
  • $\begingroup$ Unfortunately, I don't have the source for it. It was something I read about a long time ago and didn't bookmark it. $\endgroup$
    – zipline86
    Oct 15, 2019 at 9:54
  • $\begingroup$ Was it a rigorous work? Like an academic-level paper, or a highly specialized blog? Or was it something less reliable? (Just out of curiosity) $\endgroup$
    – Leevo
    Oct 15, 2019 at 10:16
  • 1
    $\begingroup$ It was something less reliable. $\endgroup$
    – zipline86
    Oct 15, 2019 at 13:08

1 Answer 1

0
$\begingroup$

I'll go through your questions one by one:

I once read about somebody who added noise to their time series before training a model. They didn't write why they did it though.

No, noise is not adding any value to your dataset. The reason is that noise, by definition, cannot be learned by any statistical/ML model.


Is this common practice?

No it's not.


why do people do it ie. to prevent over-fitting?

It's not a technique that helps you in preventing overfitting. There are several techniques to do it (ensemble modeling, cross validation), many of them depend on the kind of model you want to implement (such as recurrent dropout for RNNs).

$\endgroup$
3
  • 1
    $\begingroup$ Thanks. If there any particular reason why one would want to add noise to a time-series? $\endgroup$
    – zipline86
    Oct 15, 2019 at 9:52
  • $\begingroup$ Perhaps if you want to create some toy dataset and test some simulations? But I can't find rigorous, scientific reasons. $\endgroup$
    – Leevo
    Oct 15, 2019 at 10:34
  • $\begingroup$ Ok great, thanks for the help. $\endgroup$
    – zipline86
    Oct 15, 2019 at 13:09

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.