0
$\begingroup$

I have UsArrests dataframe and i am trying k means clustering algorithm. I ve tried with and without scaling the data . How i can decide which data i must take ? I must see the Within cluster sum of squares by cluster? With out scaling the Within cluster sum of squares by cluster is 86.5 and with the scaling is 60 % .

Improve this question
$\endgroup$
4
  • 1
    $\begingroup$ If/when this question is migrated to Cross Validated, you will still need to fill out some details. Realize that it's very difficult to say "which data" (or answer the sum-of-sq question) without knowing anything about the data. $\endgroup$
    – r2evans
    Commented Jun 29, 2021 at 19:45
  • $\begingroup$ Hello my friend. i mean which is a better number ? $\endgroup$
    – Kleanthis Mpampotsi
    Commented Jun 29, 2021 at 19:52
  • 1
    $\begingroup$ Assuming you are using the USArrests data set that is included with R, there are four variables. The means for those variables range from 7.8 (Murder) to 170.8 (Assault). Furthermore, three variables are arrest rates and one is the percentage urban population. If you use the raw data, your clusters will primarily reflect differences in Assault rate and secondarily in UrbanPop. It is not clear what research question(s) clusters based primarily on those two variables would answer. $\endgroup$
    – dcarlson
    Commented Jun 29, 2021 at 20:43
  • $\begingroup$ I think your forcing to solve the problem through clustering. I further think you want to perform unsupervised feature selection. Furthermore, accuracy is a supervised metric. It has no role to play in unsupervised problems. Search for cluster purity metric. $\endgroup$
    – mnm
    Commented Jul 1, 2021 at 9:10

0

Reset to default

Your Answer

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

Browse other questions tagged or ask your own question.