Accession Number : AD0643287

Title :   A COMPARISON OF SOME CLUSTER-SEEKING TECHNIQUES.

Descriptive Note : Technical rept., Jun-Aug 66,

Corporate Author : STANFORD RESEARCH INST MENLO PARK CALIF

Personal Author(s) : Bell,Geoffrey H.

Report Date : NOV 1966

Pagination or Media Count : 52

Abstract : Conventional multivariate statistics examine in considerable depth the significance of relationships existing in data as shown by the mean and the covariance matrix. Shortcomings of this approach are briefly discussed. 'Cluster-seeking techniques' are discussed as alternatives to conventional multivariate methods. Thirty variants of cluster-seeking techniques, the total number presently known to the author, are divided into seven categories: probabilistic, signal detection, clustering, clumping, eigenvalue, minimal mode seeking and miscellaneous. These larger classes are contrasted and, within each class, the techniques are summarized and compared. A composite technique that combines the best features of the various approaches is proposed. (Author)

Descriptors :   (*MULTIVARIATE ANALYSIS, EFFECTIVENESS), S MATRIX, SOCIAL SCIENCES, CORRELATION TECHNIQUES, PROBABILITY, INFORMATION THEORY

Subject Categories : Statistics and Probability

Distribution Statement : APPROVED FOR PUBLIC RELEASE