Skip to content
ML Academy

Introduced on day 14

k-means

A clustering algorithm that assigns each row to the nearest of k centres and moves the centres to the mean of their rows until assignments stop changing.

In detail

k-means minimizes the within-cluster sum of squared Euclidean distances. It needs k in advance, assumes compact, roughly round clusters of similar size, depends on feature scaling, can stop at a local minimum (use several starts and k-means++ seeding) and always returns k clusters, including on data with no groups.