Mining probabilistic rules using nonmonotonic rule layers

Abstract

This paper proposes a new framework for rule induction methods based on rule layers constrained by inequalities of accuracy and coverage. When the changes of accuracy and coverage are considered with an additional example, four patterns of updates of accuracy and coverage are observed and give two important inequalities of accuracy and coverage for induction of probabilistic rules. By using these two inequalities, the proposed method classifies a set of formulae into four layers: the rule layer, subrule layer (in and out) and the non-rule layer. Using these layers, updates of probabilistic rules are equivalent to their movement between layers. The proposed method was evaluated on datasets regarding headaches and meningitis, and the results show that the proposed method outperforms the conventional methods.