What is the primary goal of a clustering method such as k-means?

Clustering algorithms partition a dataset into subsets (clusters) so that members of the same cluster are more similar to each other than to members of other clusters, according to a chosen distance or similarity measure. K-means is a canonical example: the practitioner selects k, initializes k centroids, assigns each point to the nearest centroid, recomputes centroids as cluster means, and iterates until assignments stabilize or a budget is exhausted. No ground-truth class labels are required; the algorithm discovers structure in unlabeled data—hence unsupervised learning.

Supervised classification instead maps inputs to predefined categories using labeled examples; that is a different problem statement even when both produce "groups" in casual language. Regression predicts continuous targets and likewise needs labeled (x, y) pairs. Dimensionality-reduction techniques such as PCA transform the feature space to fewer coordinates that preserve variance or structure; they may be used before clustering, but they do not themselves define the clustering objective of grouping instances.

In practice, k-means assumes roughly spherical, similarly scaled clusters and is sensitive to feature scaling and the choice of k (often tuned with silhouette scores, elbow plots, or domain knowledge). Alternatives include hierarchical clustering, DBSCAN (density-based, variable cluster shapes), and Gaussian mixture models (soft membership). Underlying principle: clustering discovers groups from similarity without needing predefined class names. Best practice scales features, validates k with domain judgment and internal metrics, and does not confuse clusters with supervised labels. For the exam, map the verbs carefully: classify with labels means supervised classification; predict a number means regression; shrink dimensions means feature reduction; group by similarity without labels means clustering.