I'm planning to use z-score as a Normalization Method for a Project but I noticed if I do that then I ll have a data in the range [-1, 1] which is wierd because I have data that doesn't make sense for it to have negative values. Let's say for example speed or distance, it doesn't make sense that speed would have a negative value after normalization! is it logical to think like this or am I wrong and it is perfectly fine to use z-score even if the data will be negative and it doesn't make sense?

ps: I know velocity can be negative if we are talking about Vectors but I meant to say that I have a discrete Values for Speed or Distance a.k.a Length of something which cannot be negative.

  • $\begingroup$ This depends on the model you're using...could you add some more information regarding the project? $\endgroup$ – ec2604 Oct 14 '19 at 12:50

I think it doesn't really matter that there can be negative values after as long as you rescale your data correctly at the end.

Think about what positive/negative values mean for z-score. This has nothing to do with whether your use case (for example speed) can realistically have negative values or not. With z-score positive values simply mean that the value is above the group mean while negative values can be interpreted as the opposite.

As long as you are able to rescale after your model (to get your data back to the original interpretation) using z-score should be fine.

  • 1
    $\begingroup$ thanks for your answer, I have one question though, by rescaling here you mean to get back the real data values from the normalized values ? basically do you mean to undo the normalization? $\endgroup$ – basilisk Oct 14 '19 at 13:13
  • 1
    $\begingroup$ exactly...for example if you use z-score and then do a prediction you get scaled/normalized data out of your model. For using the data you would now have to undo the normalization on your predictions to be able to use it. i.e. if you want to apply your predicted speeds then values between -2 and +2 won't help => rescaling would do the trick $\endgroup$ – Philipp Oct 14 '19 at 15:35
  • $\begingroup$ and how about if I normalized only the inputs and kept the target as it is? in that case I should not need to rescale the predicted output of the model or am I wrong? would that also work? $\endgroup$ – basilisk Oct 14 '19 at 16:03
  • $\begingroup$ Yes, that would work $\endgroup$ – Philipp Oct 14 '19 at 17:16

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.