Unsupervised Learning — Tutorial

No labels required

Unsupervised learning finds patterns in data without any human-provided labels. The algorithm must discover structure on its own.

This is valuable because labeling is expensive and sometimes impossible (e.g. grouping customer segments, detecting anomalies with no known anomaly examples, compressing data, visualising high-dimensional spaces). The two main families are clustering (group similar points) and dimensionality reduction (find compact representations).

Clustering — finding natural groups

K-Means: assigns each point to one of K cluster centres; iterates between assignment and centre recalculation. Fast, simple, assumes convex clusters of similar size.

DBSCAN: density-based clustering; finds arbitrarily shaped clusters and labels outliers as noise. Needs no K; requires two hyperparameters: epsilon (neighbourhood radius) and min_samples.

Hierarchical clustering: builds a tree (dendrogram) of merges; lets you choose the number of clusters after the fact by cutting the tree.

Dimensionality reduction — compressing while preserving structure

PCA (Principal Component Analysis): finds orthogonal axes of maximum variance and projects data onto the top-k components. Linear; fast; interpretable. Use for preprocessing before ML or for visualisation.

t-SNE / UMAP: non-linear methods for 2D/3D visualisation of high-dimensional data (embeddings, images). t-SNE is slow on large datasets; UMAP is faster and preserves more global structure.

Rule of thumb: use PCA for preprocessing and compression; use UMAP/t-SNE for exploratory visualisation only.

Continue your research

Use these connected ModelRefs sections to compare alternatives, inspect implementation paths, and review the evidence and governance boundaries relevant to Unsupervised Learning — Tutorial.

• Browse models
• Review providers
• Inspect benchmarks
• Explore workflows
• Learn AI concepts
• Read decision guides
• Review trust and methodology
• Start a decision path