k-medoids

Algorithm

The k-medoids algorithm works like k-means, with the important difference that at any given time a cluster centre has to correspond to an input data item. On each round, the input data item that is closest to the geometric centre of each cluster is selected to be used as the new cluster centre for the subsequent round. The two subtypes, Partitioning around medoids and Voronoi iteration, are simply different ways of achieving this mathematically.

K-medoids is less sensitive to the presence of outliers than k-means.

alias
subtype
Partitioning around medoids Voronoi iteration
has functional building block
FBB_Classification
has input data type
IDT_Vector of quantitative variables
has internal model
has output data type
ODT_Vector of quantitative variables ODT_Classification
has learning style
LST_Unsupervised
has parametricity
PRM_Nonparametric with hyperparameter(s)
has relevance
REL_Relevant
uses
sometimes supports
ALG_Nearest Neighbour
mathematically similar to