Fuzzifying Hyperplanes in the Hypothesis Space
Martin Eineborg · 2002
This paper is concerned with the assumption made by inductive logic programming, that the relations in an induced hypothesis must have a precise definition. In this paper the rules of a hypothesis are made fuzzy by applying a method, originally intended for classifying uncovered examples, called rule stretching when classifying new examples. A number of experiments were performed where hypotheses were induced using the inductive logic programming system Virtual Predict. The rule stretching method was compared with classifying test examples using the most accurate rule. The encouraging result of the experiments showed that rule stretching outperforms the most accurate rule in six out of seven experiments. These keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.