Partitioning data points into subsets of objects according to their mutual "similarity," without using preexisting knowledge such as class labels.


Clustering, or cluster analysis, is a statistical technique of uncovering groups of units in multivariate data. It is separate from classification (clustering could be called "classification without a teacher"), as there is no units with known labels, and even the number of clusters is usually unknown, and needs to be estimated. Clustering is a key challenge of data mining, in particular when done in large databases.

Although there are many clustering techniques, they fall into several broad classes: hierarchical clustering (in which a hierarchy is built from each unit representing their own cluster up to the whole sample being one single cluster), centroid-based clustering (in which are units are put into the cluster nearest to a specific centroid), distribution- or model-based clustering (in which clusters are assumed to follow a specific distribution, such as multivariate Gaussian), and density-based clustering (in which clusters are obtained as the areas of the highest estimated density).

history | excerpt history