A binary classification methodology applicable to defects detection. Boosting algorithms

Laboratoire de Traitement du Signal et de Modélisation des Machines, Institut d'Études Supérieures de la Guyane, avenue d'Estrée, BP 792, 97337 Cayenne Cedex, France

Corresponding author: isabelle.mariejoseph@guyane.univ-ag.fr

Received: 27 April 2000
Accepted: 31 July 2000
Published online: 15 October 2000

Abstract

This article presents a binary classification method which is used in defects detection. It's presented as recursives “boosting” algorithms which allow us to obtain a precise discriminating function by combination of hypothesis and rules with moderate accuracy. This approach permits the study of random phenomena governed by nonparametric laws and a direct decision for the observations classification and the determination of frontiers in an observation space. The various analyses which will be developed are illustrated by simulations making it possible to evaluate the possibilities of the method.