A1629
Title: Notations for dendrograms and asymptotic properties of single-linkage clustering in high-dimensional settings
Authors: Yota Takao - Tokyo University of Science (Japan) [presenting]
Kento Egashira - Tokyo University of Science (Japan)
Abstract: Agglomerative hierarchical clustering using single linkage is analyzed for high-dimensional data randomly obtained from two or more independent populations. The aim is to clarify the relationship between dendrogram shapes formed by high-dimensional data and the parameter space of the populations under less stringent assumptions. Agglomerative hierarchical clustering is a method that groups data for which prior information such as the true number of clusters is not provided according to their similarity. Users of agglomerative hierarchical clustering can understand characteristics of the data under investigation by analyzing the resulting dendrograms. Notations representing various dendrogram shapes are introduced to facilitate discussion of dendrogram structures. Using these notations, asymptotic properties of agglomerative hierarchical clustering with single linkage are established as the data dimension diverges.
