AUTOMATIC CONSTRUCTION OF A HIERARCHICAL CLASSIFIER

D. Filbert, F. Attia, R. Jahnke

Abstract

The automatic construction of a hierarchical classifier is described. The construction process uses the same classifier to select the features, which is used later for the classification itself. The construction leads to a binary decision tree. Every node is labelled with a feature vector.<BR />The classical statistical approach to feature selection is presented. The add-on algorithm provides feature vectors of minimal length without regarding the classifier. Two applications are described and the results, reached by the classical statistical algorithm and the new hierarchical classifier, are compared.

Keywords

Decision tree, supervised learning, classification, feature evaluation

Download

IMEKO-WC-2000-TC10-P262.pdf

DOI

-

IMEKO TC

TC10 - Measurement for Diagnostics, Optimization and Control

Back to the proceedings