CODEWITHSUNDEEP
orange
taylor
taylor

Reputation: 55

orange3 clustering with k means

I am new to orange, and i am having trouble generating a scatter plot of clusters with k means. Here is a picture of the clusters I have.output of scatterplot from orange

My process so far has been uploading the data set, preprocessing by normalizing, PCA, k-means with the number of clusters as the highest silhouette score, and then doing a scatter plot and coloring by cluster. Why are my clusters overlapping? Additionally, why would there be low numbers for each silhouette score (several at ~0.19 and not one number of clusters really dominating)?

Upvotes: 0

Views: 732

Answers (1)

JanezD
JanezD

Reputation: 545

You are looking at a shadow of your data. :)

K-means uses all data, but scatter plot shows only two-dimensional projection. It is like seeing overlapping shadows of two persons, even though their bodies do not overlap in the real, 3-d world.

Similar silhouette scores for different k's simply mean that the number of clusters is not clearly defined.

Upvotes: 1

Related Questions