INCOMPLETE PRIOR INFORMATION IN A CLASSIFICATION PROCEDURE.
Abstract
The classification of specimens into one of N populations is considered when the observation on each specimen is a multivariate discrete random variable and when incomplete prior information is available concerning the occurrence of specimens from the various populations. The T-minimax decision criterion is applied and it is shown that an optimal classification procedure is the solution of a finite matrix game. The consistency of the T-minimax procedure is demonstrated when certain population probabilities are estmated by sample proportions. A medical example is considered. (Author)
Document Details
- Document Type
- Technical Report
- Publication Date
- Jul 01, 1970
- Accession Number
- AD0712031
Entities
People
- Myles Hollander
- Ronald H. Randles
Organizations
- Florida State University