Paper
W. Van Laer, S. Dzeroski, and L. De Raedt, "Multi-class problems and
discretization in ICL " in Proceedings of the MLnet Familiarization
Workshop on Data Mining with Inductive Logic Programming (ILP for
KDD), ed. B. Pfahringer en J. Fuernkranz, juli 1996, pp. 53-60
[ps format]
Abstract
Handling multi-class problems and real numbers is important in practical
applications of machine learning to KDD problems. While attribute-value
learners address these problems as a rule, very few ILP systems do so.
The few ILP systems that handle real numbers mostly do so by trying out
all real values that are applicable, thus running into efficiency or
overfitting problems. This paper discusses some recent extensions of ICL
that address these problems.
ICL, which stands for Inductive Constraint Logic, is an ILP system
that learns first order logic formulae from positive and negative examples.
The main charateristic of ICL is its view on examples. These are seen as
interpretations which are true or false for the clausal target theory
(in CNF).
We first argue that ICL can be used for learning a theory in
a disjunctive normal form (DNF). With this in mind, a possible solution
for handling more than two classes is given (based on some ideas from CN2).
Finally, we show how to tackle problems with continuous values by adapting
discretization techniques from attribute value learners.