Rule induction-machine learning techniques

With artificial neural networks (ANN) now approaching the exploitable stage and their use being promoted under the DTI\´s "Neural Computing: Learning Solutions Campaign", it is an appropriate time to consider an alternative set of machine learning techniques in which production rules are generated from sets of data. The techniques used in rule induction offer a powerful alternative approach to ANN but, since both deal in essence with the production of a classifier from a set of example data, it is not surprising that both are being successfully applied to the same types of problems. In turn, the drawbacks of each of the methods are similar, although practitioners of each approach argue that their method has advantages over the other. Much is currently being heard about the benefits of ANN, and this article is intended to redress the balance in a small way by describing rule induction techniques and how they can be applied. Inevitably some advantages over ANN are identified, but it is expected that the ANN community will have adequate answers to any criticism.<>