Metric Considerations in Cluster Analysis.
Abstract
A variation of the 'k-means' method of cluster analysis is described which is designed to take into account and profit from the possibility that the separate clusters resemble samples from multivariate normal distributions with substantially different covariance structures. These covariance structures determine metrics which can be updated recursively. (Author)
Document Details
- Document Type
- Technical Report
- Publication Date
- Sep 20, 1970
- Accession Number
- AD0714810
Entities
People
- Herman Chernoff
Organizations
- Stanford University