Self-expanded clustering algorithm based on density units

An efficient self-expanded clustering algorithm based on density units (SECDU) is presented. The whole data space is divided into several density units equally. Each data point is put into a density unit according to the data point possition. The area with the highest data density is the starting point of clustering and it is expanded to the low-density area. The whole process will not stop until densities of all clusters reduce to the threshold set in advance. By compressing data into data units, SECDU can cluster large dataset at a high speed without destroying distribution feature.