Data Science Glossary - K

Cluster analysis.  K-Means produces poor results,
while EM did better. The visualization was
generated using ELKI.  (Image credit: Wikipedia)

K-Means Clustering

is a clustering method is the most basic clustering algorithm. This method has complexity of O(N). But it can converge to a locally optimal clustering solution, it is sensitive to outliers, noise only works on convex clusters and does not assist in selecting the correct number of clusters. K-means is a hard selection clustering method - for the soft selection method it is possible to use the expectation-maximization algorithm for mixtures of Gaussian distributions.

K-Medoids Clustering

is a clustering method based on K-Means which uses representatives for the means rather than computed geometric center of the cluster. It is more robust to noise and outliers than K-means.

K-Medians Clustering

is a clustering method based on K-Means, it selects the median point geometrically using the Manhattan distance and will provide more compact clusters.

K-Modes Clustering

is a clustering method based on K-Means which can be used to cluster categorical data using a number of multidimensional features.

Kernel K-Means Clustering

is a clustering method based on K-Means using the Kernel trick to solve cases where clusters are non convex. However this method has complexity of $O(N^2)$.