Feature Selection Algorithms Using Non-Redundant Thresholded Measures.
Abstract
A new feature selection method, the threshold selection algorithm, is presented and compared with sequential selection and rejection algorithms. This algorithm assumes a measure of feature discrimination exists and provides a set of threshold parameters, associated with class pairs, which are dynamically variable. The basis of comparison of the algorithms is a pattern recognition system operating on hand-printed alphabetic characters. The threshold selection algorithm provides improvement (in terms of system error rate) over sequential selection and rejection. Finally, a modified threshold selection algorithm with a redundancy measure is described which exhibits a considerable improvement in performance.
Document Details
- Document Type
- Technical Report
- Publication Date
- Dec 01, 1974
- Accession Number
- ADA004965
Entities
People
- Allen R. Hanson
- Edward G. Fisher
- Edward M. Riseman
Organizations
- University of Massachusetts Amherst