- clustering algo where each point belongs in only 1 cluster
- clusters are always circular
- steps
-
- select a number of classes/groups to use and randomly initialize their respective center points
-
- classify each datapoint as the cluster with the nearest group cluster.
- nearest depends on a distance metrics
-
- recompute the group center by taking the mean of all vectors in the group
-
- repeat 2 & 3 until convergence