Supervised learning uses labeled examples (known outcomes). Unsupervised learning finds structure without labels—clusters, segments, anomalies.
Supervised tasks
- Classification — spam vs not, churn yes/no
- Regression — predict revenue, temperature, ETA
Labels must be correct and timely—garbage labels produce garbage models.
Unsupervised tasks
- Clustering — customer segments, document themes
- Dimensionality reduction — compress features for visualization
- Anomaly detection — rare fraud or equipment failure patterns
Choosing approach
If stakeholders need actionable groups and labels are expensive, clustering plus domain review may suffice. If decisions need probability of churn, supervised classification fits.
Semi-supervised and beyond
Large unlabeled data plus small labeled set—advanced topic. Start with clear supervised or unsupervised framing in interviews.
Important interview questions and answers
- Q: Supervised vs unsupervised?
A: Supervised uses labeled y; unsupervised discovers patterns without y. - Q: Clustering always useful?
A: Clusters need business interpretation—may not align with actionable segments.
Self-check
- Give one classification and one regression example.
- What is clustering used for?
- When prefer supervised over clustering?
Tip: Clustering needs business interpretation—not just k choice.
Interview prep
- Clustering?
Unsupervised grouping without labels.