Induction of relational Fril rules

Abstract

We propose an, approach to extend inductive logic programming (ILP) to cater for uncertainties in the form of probabilities and fuzzy sets. A corresponding decision tree induction algorithm that induces Fril (a support logic programming language) classification. Rules involving both forms of uncertainties is also described. This algorithm iteratively builds decision trees where each decision tree consists of one branch. This branch is directly translated into Fril rules that explain a part of the problem. The work presented focuses on propositional representations for both the input data values and the learned models. The approach is illustrated on the Pima Indian dataset. Finally an overview of the current work is given which deals with improving the algorithm with a new method for the calculation of support pairs and also with a new, user-independent stopping criterion for adding literals to the body of a rule.