3
$\begingroup$

In my machine learning class these two methods were discussed and mentioned that both should be used. I have a couple questions about this:

1) Can I mix and match these two approaches? e.g. Feature Scale x1 and Mean Normalize x2?

2) How do you determine which of these options to apply? It seems that either could accomplish the task of increasing your convergence rates... I suppose you just need to know your data set to understand which will reliably reduce your values while leaving as few outliers as possible?

$\endgroup$
1
$\begingroup$

I agree with the existing answer that feature scaling is a superset into which techniques like mean normalization, residual normalization, etc falls under.

So, assuming that by feature scaling, you mean the techniques other than mean normalization, I would attempt to answer your questions:

1) Can I mix and match these two approaches? e.g. Feature Scale x1 and Mean Normalize x2?

In most cases No. Generally, only one normalization technique is used and it pretty much suffices the need. In addition to that argument, it should also be noted that any normalization technique introduces duplication in the data records (not necessarily redundant duplication).

So, pretty much a single normalization technique would suffice most of the times.

2) How do you determine which of these options to apply? It seems that either could accomplish the task of increasing your convergence rates... I suppose you just need to know your data set to understand which will reliably reduce your values while leaving as few outliers as possible?

Yes, you are right. The selection of the technique depends on the data. And the feature scaling (and normalization) process comes under the process of data cleaning. So, it is done immediately after the selection of the relevant data for the analytics process.

| improve this answer | |
$\endgroup$
2
$\begingroup$

Mean normalization is a form of feature scaling, so these are not really two different approaches. Feature scaling is just a more general term. What kind of feature scaling, e.g. mean normalization, you need to use ultimately depends on the data.

| improve this answer | |
$\endgroup$
  • $\begingroup$ Makes sense... but what about how to choose what method to feature scale and can each feature be manipulated differently than any other feature? $\endgroup$ – Dave Nov 20 '15 at 21:58
  • 1
    $\begingroup$ That entirely depends on the data. $\endgroup$ – Felix Darvas Nov 20 '15 at 22:29

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.