Browse Prior Art Database

Procedure for Partitioning Classes Into Sub-Classes for Classification Purposes

IP.com Disclosure Number: IPCOM000037116D
Original Publication Date: 1989-Nov-01
Included in the Prior Art Database: 2005-Jan-29

Publishing Venue

IBM

Related People

Authors:
Bahl, LR deSouza, PV Mercer, RL Brown, PF [+details]

Abstract

The problem of classification is to allocate a given parameter vector to one of several predetermined classes. The most prominent approach to classification is to construct linear classifiers which partition the parameter space into piecewise linear cells in 1-1 correspondence with the classes. A parameter vector may then be classified by determining in which cell it is located. This approach may not succeed, however, if any of the classes is non-homogeneous, i.e., if any of the classes is actually a super-class consisting of several unrelated sub-classes. This invention provides a means of determining how many sub-classes a given class should be decomposed into. The sub-class locations in the parameter space are then determined via standard cluster-analytic techniques.