Estimate K-Means inertia, the within-cluster sum of squares, from sample points and centroids, to compare clustering configurations.
Inertia 越小簇内越紧密。
| Item | Formula / Note | Source / Standard |
|---|---|---|
| Objective | k-means minimises the sum of squared distances from samples to their centroids | Lloyd 1982, IEEE Trans. IT |
| Iteration | alternate assignment of samples and centroid update until centroids stabilise | MacQueen 1967, Berkeley Symposium |
| Local optimum | results depend on initialisation so multiple restarts are used to pick the best | Arthur & Vassilvitskii 2007, SODA |