single-jc.php

JACIII Vol.3 No.5 pp. 357-367
doi: 10.20965/jaciii.1999.p0357
(1999)

Paper:

Fuzzy Rule Acquisition from Trained Artificial Neural Networks

Peter Geczy, Shiro Usui

Department of Information and Computer Sciences Toyohashi University of Technology Hibarigaoka, Toyohashi 441-8580, Japan

Received:
May 7, 1999
Accepted:
September 20, 1999
Published:
October 20, 1999
Keywords:
Logical formalism, Rule extraction, Fuzzy rules, Classification, Partitioning, Mapping, Training, MLP neural networks
Abstract

We approach the problem of rule extraction in its primary form. That is, given a trained artificial neural network, we extract rules classifying data set as correctly as possible. Attention is oriented toward extraction of fuzzy rules. The choice of fuzzy rules underlines the aim of balancing rule comprehensibility and complexity. To achieve higher comprehensibility of extracted rules, the formulated theoretical material is an extension of crisp rule extraction 1). A rule extraction algorithm is introduced. The presented algorithm for fuzzy rule extraction implies from the derived theoretical results rather than from heuristics. The rule extraction algorithm incorporates a ’built-in’ rule simplification mechanism. This feature is beneficial in cases when trained neural network structure is overdetermined for a given task. The rule extraction algorithm is experimentally demonstrated. Demonstrations incorporate both structure modification training and fixed structure training.

Cite this article as:
Peter Geczy and Shiro Usui, “Fuzzy Rule Acquisition from Trained Artificial Neural Networks,” J. Adv. Comput. Intell. Intell. Inform., Vol.3, No.5, pp. 357-367, 1999.
Data files:

*This site is desgined based on HTML5 and CSS3 for modern browsers, e.g. Chrome, Firefox, Safari, Edge, Opera.

Last updated on Mar. 05, 2021