Adaptation of the ISODATA clustering algorithm for vector supercomputer execution

Abstract

Cluster analysis is an interdisciplinary study which involves the grouping of similar objects based on their measured attributes. The purpose of a cluster analysis is to investigate the structure and organization of the objects being studied. A description is given of the adaptation of the ISODATA clustering algorithm for vector supercomputer execution. On the CYBER 205, the algorithm runs 30 times faster than the original algorithm on the CYBER 205 using full automatic vectorization and 300 times faster than on a VAX 11/780. The major source of improvement over automatic vectorization is achieved by reorganizing the data structures used by the program. The modified algorithm yields increased performance on any vector computer.<>