Bootstrapping Cluster Analysis Solutions with the R package ClusBoot

Finding true clusters in an unsupervised setting is a difficult problem. In most cases a data set can be clustered into a specific number of clusters whether this supports the underlying structure of the data or not. The package ClusBoot uses a bootstrap analysis of any clustering algorithm to provide its user with some measures of the stability in the clustering solution. Observations that cluster together repeatedly over many bootstrap replications can be considered similar enough to be grouped into a cluster while observations that only cluster together by chance indicates a lack of true grouping structure. The package performs the bootstrap analysis and provide the user with summary measures in the form of a bootstrap-silhouette plot and graphical visualisation to assess the stability of the clustering solution.