1. Home
  2. Glossary
  3. Unsupervised learning
AI glossary · Foundations

Unsupervised learning

Unsupervised learning: Unsupervised learning is machine learning on data with no correct answers attached. The model finds structure on its own, grouping similar customers, spotting unusual transactions, or clustering tickets by topic; a human then decides what the groups mean.

Labeled data is expensive and often unavailable, so a whole branch of machine learning works without it. Give an unsupervised model your customer records and it will sort them into clusters of similar behavior. Give it a year of server logs and it will flag the hours that look nothing like the rest. It does not know what a "high-value customer" or an "outage" is; it only knows what is similar and what is unusual.

The common techniques have plain-English jobs. Clustering groups things by similarity. Dimensionality reduction squeezes hundreds of columns down to a few that carry most of the signal, often so the data can be charted. Anomaly detection isolates the outliers. Association mining finds items that show up together, the classic "people who bought this also bought" pattern.

The catch is interpretation. The model hands you cluster 3; you have to look inside and decide it is "small accounts that order weekly and never call support". Clusters can also be artifacts of how the data was formatted rather than anything real. Treat unsupervised output as a set of hypotheses to check with someone who knows the business, not as findings.

Example at work

A marketing analyst runs clustering on two years of purchase history and gets five customer groups. Three match segments the team already used. One is a surprise: customers who buy only during promotions and return a large share of what they order. That group gets excluded from the next discount campaign, which improves margin without hurting overall sales.

Why it matters

Unsupervised learning is how you find patterns you did not know to look for, which is often more valuable than confirming what you expected. Knowing that its output needs human interpretation keeps you from mistaking a cluster for a conclusion.

Related terms