Classical Machine Learning

Clustering & Dimensionality Reduction

Clustering groups examples by similarity; dimensionality reduction compresses many features into fewer coordinates. Both are powerful for exploration and representation, but neither gives truth by itself.

K-means finds compact clusters around centroids
Hierarchical clustering shows nested structure
PCA finds directions of maximum variance
t-SNE/UMAP-style projections are visualization tools, not proof
Embedding models are learned dimensionality reduction
Unsupervised outputs need external validation

What each method is for
Method	Best use	Watch out
K-means	Fast grouping around centroids	Must choose k; scale-sensitive
Hierarchical clustering	Exploring nested group structure	Expensive at scale
PCA	Linear compression / noise reduction	Variance is not meaning
t-SNE / UMAP	Visualization	2D shapes are easy to overread
Embeddings	Semantic retrieval / recommendation	Similarity depends on training objective

Unsupervised workflow

The algorithm proposes structure; validation decides whether the structure matters.