Normalisation transform data into a range: $$X_i = \dfrac{X_i - Min}{Max-Min}$$

Practically, I found out that the model doesn't generalise well when using normalisation of input data, instead of standardisation (another formula shown below).

Before training a neural net, data are usually standardised or normalised. Standardising seems good as it makes the model generalise better, while normalisation may make the model not working with values out of training data range.

So I'm using standardisation for input data (X), however, I'm confusing whether I should standardise the expected output values too?

For a column in input data: $$X_i = \dfrac{(X_i - Mean)}{Standard\ Deviation\ of\ the\ Column}$$

Should I apply this formula to the expected output values (labels) too?

  • $\begingroup$ What is the difference between "standardization" and "normalization"? I think these terms would refer to the same concept and, from your equation, it seems like you're referring to some form of "normalization". Can you clarify that? Also, maybe you should describe the type of task you're trying to solve: is it classification or regression? How do your labels look like? Are they integers or floating-point numbers? Do they have a range? $\endgroup$ – nbro Jul 25 '20 at 2:46
  • $\begingroup$ normalisation: values are scaled into a range [a,b] that is: x = (x-min)/(max-min); and standardisation is x = (x-mean)/(sqrt(sum((x-mean)^2)/n)); they r different concepts $\endgroup$ – datdinhquoc Jul 26 '20 at 3:20

It depends, as mentioned in comments, on your model and labels. For example how would you use standardisation on multi classification problem?

Generally, standardisation is more flavourable for input data as its mean is around 0.

I assume you have a regression model and in that case using standardisation could be better than normalisation.


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.