A Modular Neural Network for Vague Classification

  • Authors:
  • Gasser Auda;Mohamed S. Kamel

  • Affiliations:
  • -;-

  • Venue:
  • RSCTC '00 Revised Papers from the Second International Conference on Rough Sets and Current Trends in Computing
  • Year:
  • 2000

Quantified Score

Hi-index 0.00

Visualization

Abstract

A modular neural network classifier design is presented. The objective behind the design is to enhance the classification performance of conventional neural classifiers according to two criteria, namely, reducing the classification error, and allowing vague/boundary classification decisions. The proposed model uses an unsupervised network to decompose the classification task over a number of neural network modules. During learning, every module is trained using samples representing the other modules, and modules are trained in parallel. After the training phase, every module inhibits or enhances the responses of the other modules by "voting" for the existence of the input within their decision boundaries. If the result of the majority vote is a "tie", then the sample is classified as a vague class (or boundary) between the (two or more) classes that have the tie. The proposed classifier is tested using a two-dimensional illustrative benchmark classification problem. Results are showing an enhancement in the classification performance according to the above two criteria.