Hi, this article was clear and concise!

I've been working with features of Twitter data along with derived features like sentiment scores. I ran into scaling issues doing 3-d plots with attributes with very different dimensions and distributions. I did 'absolute' scaling first, but in many cases that was only a marginal improvement.

The 3rd and 4th methods you describe, z-scores and robust scaling, solved my issues. I think its always good to explore your data and understand the presence of outliers as they can do screwy things to analysis. Your explanation of how robust scaling avoids being swayed by outliers applies to a lot of real-world datasets! Good Article!