LEAST SQUARES FIT TO THE OPTIMUM (BAYES') DECISION SURFACE.
Abstract
A procedure for the determination of a high-order polynomial decision surface is presented. The procedure is based on the least squares fitting of polynomials to the optimum (Bayes) decision surface. The weighting function required for the least squares procedure is constructed so that the resulting polynomial decision surface will minimize an approximation to the risk of misclassification. The least squares procedure is compared in both theory and simple application to a similar procedure due to Specht which is based on a Taylor series approximation to the Bayes solution. It is shown that the least squares procedure yields significantly superior results in some situations. (Author)
Document Details
- Document Type
- Technical Report
- Publication Date
- Feb 09, 1968
- Accession Number
- AD0671908
Entities
People
- A. R. Williams
- D. B. Brick
- W. N. Furey