Note on Probability of Error on Design Set.
Abstract
SYSTEM DESIGN IS TYPICALLY DONE USING DATA SAMPLES FROM THE PROBLEM CLASSES. A problem of interest is an estimate of the pattern recognizer performance based upon experimental results. This problem under the assumption of normal class conditional distributions of known covariance was investigated by Foley (IEEE Transactions of Information Theory, 'Considerations of Sample and Feature Size,' page 618, September 1972). The note addressed the same problem under the weaker condition that known covariance is replaced by sample estimated covariance. Estimated performance is discussed in terms of data dimensionality and number of samples. (Author)
Document Details
- Document Type
- Technical Report
- Publication Date
- Apr 01, 1973
- Accession Number
- AD0761794
Entities
People
- Kishan G. Mehrotra
Organizations
- Syracuse University