Review & Self Study — A Cluster Is Not a Type

Clustering can reveal useful structure without target labels, but the analyst still chooses the representation, distance assumptions, and often the number of groups.

Review questions

1. What is the difference between supervised and unsupervised learning?
2. What two steps repeat in k-means?
3. What does a centroid represent?
4. Why can feature scaling change the grouping?
5. Why can an outlier move a centroid?
6. What is wrong with automatically naming a cluster as a real-world type?

Compare:

Cluster 2 contains machines with higher vibration under this representation.

Cluster 2 is the failing-machine class.

The first sentence describes the observed grouping. The second invents a target label that was never provided or validated.

Self study

• Sketch the same six points under k = 2 and k = 3.
• Find a clustering application where the groups are used for exploration rather than permanent classification.
• Describe one feature you would remove because it encodes an irrelevant shortcut.
• Re-run the playground with an outlier and explain whether you would investigate, remove, or preserve that point before knowing more.