Refinement of temporal constraints in fuzzy associations

The objectives of hypothesis refinement in knowledge discovery are to produce rules that more accurately model the underlying data while maintaining rule interpretability. In this paper we introduce two refinement strategies for association rules with fuzzy temporal constraints. Disjunctive generalization produces more general rules by merging adjacent constraints within a partition of the window of temporal relevance. Temporal specification uses linguistic hedges to reduce the duration of a constraint to better model the distribution of examples. Both types of refinement produce rules expressible using the linguistic terms of the original rules. The acquisition of the information needed to perform the refinements is incorporated into a general algorithm for determining the number of examples and counterexamples of rules with fuzzy temporal constraints.