SYMBOLIC RULE EXTRACTION FROM NEURAL NETWORKS - AN APPLICATION TO IDENTIFYING ORGANIZATIONS ADOPTING IT

Interest in the application of neural networks as tools for decision s upport has been growing in recent years. A major drawback often associ ated with neural networks is the difficulty in understanding the knowl edge represented by a trained network. This paper describes an approac h that can extract symbolic rules from neural networks. We illustrate how the approach successfully extracted rules from a data set collecte d from a survey of the service sectors in the United Kingdom. The extr acted rules were then used to distinguish between organizations using computers from those that do not. The classification scheme based on t hese rules was used to identify specific segments of a market for prom oting adoption of information technology The extracted rules are not o nly concise but also outperform discriminant analysis in terms of pred ictive accuracy. (C) 1998 Elsevier Science B.V. All rights reserved.