1) Every observation is assigned to a cluster in both K-means and hierarchical clustering, but this may not be the best option. Both methods force every observation into a cluster, but are also sensitive to outliers and may create distorted clusters as a result.

2) Clustering is not very robust to disturbances in the data. For instance, taking out a subset of n observations may cause undesired changes between the original cluster and resulting cluster.


University of Michigan - Ann Arbor

Some decisions must be made before performing K-means and hierarchical clustering. With these methods there is no single right answer and any solution that exposes some interesting aspects of the data should be considered.

Practical Issues in Clustering

The following is a decision that must be made before performing K-means clustering:
How many clusters should we look for in the data?

Issues in regular K-means clustering

The followings are decisions that must be made before performing hierarchical clustering:

   1. What dissimilarity measure should be used?
    2. What type of linkage should be used?
   3. Where should we cut the dendrogram in order to obtain clusters?


Learn Before

Related