1
$\begingroup$

I have two input variables revenue and age. Am trying to find different bins within that variables.

For ex: I have revenue and age.

I see that my revenue data is skewly distributed and regular methods like quantiles, binning etc cannot be applied due to skewness (and gives misleading results).

Is it a good practice to scale/normalize the 1d data before we apply techniques like jenks natural breaks??

Or we should standardize/nornalize only for k-means multivariate clustering algorithm?

Improve this question
$\endgroup$

1 Answer 1

Reset to default
1
$\begingroup$

Note that Jenks natural breaks optimization is 1D k-means, and scaling and shifting is done typically for bringing different features into one scale. If there is only one feature, shifting and scaling won't change the resulting clusters if same (shifted and scaled as well) random initialization is used. But, you could apply other transformations, such as log transformation, to battle with skew and see if it helps with your cluster assignments.

Improve this answer
$\endgroup$
4
  • $\begingroup$ nice. Then can I check whether it is incorrect to do multiple 1d clustering/1d-kmeans instead of single multivariate clustering? you can find more info here - stats.stackexchange.com/questions/576980/… $\endgroup$
    – The Great
    Commented May 29, 2022 at 6:41
  • $\begingroup$ Additionally, is it good to apply a stack of transformers? Like 1st apply log transformation, on top of that I apply sqrt transformation, later I apply box transformation etc? Can I keep applying until I remove skewness to the best way possible? $\endgroup$
    – The Great
    Commented May 29, 2022 at 6:47
  • $\begingroup$ You can apply transformations any way/order you like. Note though that after log transformation, there maybe negative values, and sqrt may not be appropriate. Checking 1d vs nd clustering is a post-analysis and decoupled from what you do here. $\endgroup$
    – gunes
    Commented May 29, 2022 at 6:51
  • $\begingroup$ Understand. yes, I meant it would be helpful if I could have your views on 1d vs nd clustering post that I linked. $\endgroup$
    – The Great
    Commented May 29, 2022 at 7:05

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.