Definition
Unsupervised learning works with unlabeled data. The algorithm looks for groups, outliers, or lower-dimensional structure without being told the "correct" output for each example.
Common techniques include clustering (k-means), dimensionality reduction (PCA), and some forms of representation learning.
In simple terms
Supervised learning is sorting mail with labeled folders. Unsupervised learning is dumping letters on a table and noticing natural piles by handwriting or stamp style — no labels needed.
Where you see it
- Customer segmentation from purchase behavior.
- Topic discovery in large Somali text corpora.
- Anomaly detection in network logs.
How it works
1.Collect unlabeled data
Vectors, embeddings, or tabular features without targets.
2.Choose an algorithm
Cluster, reduce dimensions, or learn representations.
3.Interpret results
Inspect clusters or components — human judgment still matters.
Why it matters
- Most real-world data is unlabeled; unsupervised methods unlock structure before costly annotation.
Often confused
Unsupervised means no human involvement.
Humans still choose features, parameters, and how to interpret discovered structure.