The divisive clustering approach begins with a whole set composed of all the data points and divides it into smaller clusters. This can be done using a monothetic divisive method. But
what is method of monothetic divisive?
Let’s try to understand it by using the example from the agglomerative clustering section above. We consider a space with six points in it as we did before.
We name each point in the cluster as ABCDEF.
Here, we obtain all possible splits into two clusters, as shown.
For each split, we can compute cluster sum of squares as shown:
Next, we select the cluster with the largest sum of squares. Let’s assume that the sum of squared distance is the largest for the third split ABCDEF. We split the ABC out, and we’re left with the DEF on the other side. We again find this sum of squared distances and split it into clusters, as shown.
You can see the hierarchical dendrogram coming down as we start splitting everything apart. It continues to divide until every data point has its node or until we get to K (if we have set a K value).