K-means clusters data based on the distance of the data point to a centroid.

K-means requires knowing how many clusters the result will have.

The k centroids are randomly initialized, the points are grouped to the nearest centroid, the points are averaged and the centroid is moved to this new location. These steps are repeated until a stopping condition has been reached.

The features need to be normalized before using them as an input, as it affects the performance of K-Means.

K-Means does not perform well with non-spherical shapes, unless the data are appropriately transformed beforehand.