The Clustering Methods Based on the Most Similar Relation Diagram

摘要

The MSRD (Most Similar Relation Diagram) of a dataset is a diagram in which each datum is linked to its MSD (Most Similar Data) and each MSRG (Most Similar Relation Group) is linked to its MSG (Most Similar Group). In essence, the MSRD of a dataset is the visualization of the most similar relations of the data. A clustering method based on the MSRD usually includes two stages: drawing the MSRD of the dataset and selecting some links in the MSRD to cut off. In this paper, we present a package of methods for the later stage and prove that the clustering based on the MSRD can detect both the spherical and non-spherical clusters simultaneously and has the capacity to distinguish the linearly inseparable clusters in high dimensional datasets.