Clustering

What Clustering is

Clustering imposes no predefined categories. The algorithm proposes groupings, and a human decides whether those groupings mean anything useful.

How it works

k-means iteratively assigns points to the nearest centroid and recomputes centroids; hierarchical methods build a tree of merges; DBSCAN grows clusters from dense regions and leaves outliers unassigned. Distance metric and feature scaling change results dramatically.

Why it matters

It is the default first move for making sense of an unfamiliar dataset, and it powers customer segmentation work across marketing and product analytics.

Common uses

• →Customer and behavioural segmentation
• →Document topic grouping
• →Image colour quantisation
• →De-duplication