Although not appropriate to discuss fully here, we
believe each of these different kinds of rule sets have
something potentially interesting and important to
say about the instances. The last rule set is, of course,
monotonic with the addition of new instances, providing
that ambiguity is not created. We thought it
interesting to juxtapose Circle’s sets of classifiers to
those generated by ID3[7] a decision tree based on
information gain (entropy) and association rules[8].
Observe there are some differences, not the least of
which is presentation. Certainly how the classifier is
to be used would determine which, if any, are deemed
best. In our opinion, using several classifiers is probably
the best approach to most problems.