K‑Means partitions observations into K clusters by minimizing within‑cluster variance. Proper scaling and K selection are critical.
Preparation
- Standardize features; optional PCA for decorrelation or visualization
- Handle outliers that can distort centroids
Choosing K
- Elbow (inertia) and silhouette score across candidate K
- Stability across random inits and subsamples
Outputs
- Cluster labels and centroids
- Profile tables: means, counts, and top differentiating features
- Silhouette plots and separation visuals
When Not to Use
If clusters are non‑spherical or uneven density, consider DBSCAN/HDBSCAN or Gaussian Mixtures.