K-means Cluster not delineatable

Hi everybody, since I am quite new tho Data Science and Knime i was wondering if somebody could help me:

I did a k-means cluster analysis for a data set. However, the clusters are very poorly delineated from each other, instead they are rather mixed when i color the points and display them in a scatter plot. How can i imporve that?

Hi @rueschsa,

it is quite hard to tell without knowing your data a bit better - could you share some (dummyfied) data and/or an associated workflow? Did you check out this example from the Hub already, maybe there is something you can build on?

Since the k-Means nodes settings are rather limited themselves, you could also try the Optimized K-Means (Silhouette Coefficient) Component and see if there are some settings that work well for you.

Best, Lukas

2 Likes

As already suggested the K value could be not optimised. Beside this the cluster could actually be separated but in a higher dimensional space which is not visible in a scatterplot

4 Likes

Thanks for your answer. However, i am legally not allowed to share the data.
As for the optimized K-means silhuette coefficient- I can’t finde it in my node repository and drag and drop does not work. What should I do?
Thanks in advance!

File → Install KNIME Extensions

Hello @rueschsa,

Optimized K-Means (Silhouette Coefficient) is a Component. To add Component to your workflow/repository you need to go to KNIME Hub (use link @LukasS shared in his previous reply) and either do Drag&Drop or download it using icon on the right.

HubComp

For more about Components see here:
https://docs.knime.com/latest/analytics_platform_components_guide/index.html

Br,
Ivan

3 Likes

This topic was automatically closed 182 days after the last reply. New replies are no longer allowed.