Lesson 4 of 10
Unsupervised learning: discover structure
Unsupervised learning looks for patterns in data without a supplied answer for every example. Clustering can group similar customers or documents. Dimensionality reduction can summarize many measurements using fewer values for exploration or visualization.
A cluster is a mathematical grouping—not a proven real-world category. Review examples from each group, test whether the grouping is stable, and avoid assigning sensitive meanings without evidence. Anomaly detection can flag unusual records for investigation, but unusual does not always mean wrong or fraudulent.
When to use it
- Explore a large dataset before deciding what labels are useful.
- Group similar items to help people browse or prioritize them.
- Find records that differ substantially from the normal pattern.
Try it: Describe a dataset that could be clustered. What decision might the groups support, and how would a person check whether the groups are meaningful?