Definition
Clustering algorithms partition data into groups of similar items. K-means, hierarchical clustering, and DBSCAN are common choices depending on shape and noise.
Cluster quality is judged by cohesion, separation, and domain usefulness — not a single universal accuracy score.
In simple terms
Clustering is organizing a messy desk into piles of similar papers without anyone telling you the folder names in advance.
Where you see it
- Grouping users by behavior for product insights.
- Discovering dialects or topics in unlabeled text.
- Image grouping before labeling.
How it works
1.Represent items
Features or embeddings in a vector space.
2.Assign groups
Algorithm places nearby points into the same cluster.
3.Interpret
Inspect centroids or exemplars to name each group.
Why it matters
- Clustering turns raw data into discoverable structure before expensive annotation.
Often confused
The number of clusters is always known.
Choosing k is often an open problem — try several values and validate with domain experts.