Finding structure without labels.
WHAT IT IS
Training on data with no correct answers provided. The system finds structure itself.
WHAT IT DOES
- Clustering: grouping similar items
- Dimensionality reduction: simplifying complex data while keeping what matters
- Anomaly detection: identifying items unlike the rest
WHERE IT IS USED
Customer segmentation, grouping buyers by behaviour Detecting unusual transactions Organising large document collections Recommendation systems
THE ADVANTAGE
No labelling required, which removes the main cost of supervised learning.
THE LIMITATION
The structure it finds may not be the structure you care about.
A clustering algorithm groups customers by whatever pattern is strongest in the data, which may be irrelevant to your business question.
INTERPRETING THE RESULT
Requires human judgement. The system produces groups; you decide whether they mean anything.
WHEN TO USE IT
Exploring data you do not yet understand When labels do not exist and cannot easily be created
