Question: Why we don't use standardization in feature scaling?

Course Q&A Machine Learning Specialization Supervised ML: Regression and Classification
1

Hi all,

I would like to ask a question regarding feature scaling, I know there are many methods to do it, and in the course only 2 are mentioned

1 - xj / Max(X) || Result will be between 0 and 1
2 - (xj−μj) / (xj range) || Result will be between −1 and 1

But both methods are depending on max and range (max - min) so so they might be affected by outliers of the data, and as far as I read, the best way to avoid outliers in feature scaling is to use Standardization
3 - (xj - μj) / stdv(X) || Result: no specific range

So I need to know why this method was not mentioned, is that related its’ result? should the result be in specific range in order to avoid any issue in the models? and if yes, what are the potential issues that might happen?

Thanks for your time and great effort :smiley:

2

Hello @Soad_Afify,

Thank you for bringing up the z-score normalization. It was introduced in course 1 week 2 Video “Feature scaling part 2”. Please check it out.

Screenshot from 2022-07-25 06-04-25
Screenshot from 2022-07-25 06-04-251013×666 129 KB

Cheers,
Raymond

3

Thanks for the fast reply … I will review this video for sure :blush:

4

You are welcome @Soad_Afify, and have a good day!

Raymond