Neuro-IG: A Hybrid System for Selection and Elimination of Predictor Variables and non Relevant Individuals

Article type: Research Article

Authors: Atmani, Baghdad^; | Beldjilali, Bouziane

Affiliations: Leibniz Laboratory, IMAG, 46 Av Felix Viallet, 38031 Cedex Grenoble, France, e-mail: baghdad.atmani@imag.fr | Department of Computer Science, Faculty of Science, University of Oran, BP 1524 El M'Naouer 31000 Oran, Algeria, e-mail: baghdad.atmani@univ-oran.dz, bouziane.beldjilali@univ-oran.dz

Abstract: In this article we present the general architecture of a hybrid neuro-symbolic system for the selection and stepwise elimination of predictor variables and non-relevant individuals for the construction of a model. Our purpose is to design tools for extracting the relevant variables and the relevant individuals for an automatic training from data. The objective is to reduce the complexity of storage, therefore the complexity of calculation, and to gradually improve the performance of ordering, that is to say to arrive at a good quality training.

Keywords: hybrid system, neural network, automatic training, pruning, symbolic system, rule extraction

Journal: Informatica, vol. 18, no. 2, pp. 163-186, 2007