Heuristic functions for learning fuzzy conjunctive rules

When learning classification rules, many possible antecedents for a rule exist. These antecedents are usually in the form of a conjunction, and need to be evaluated for their classification performance on a training set of instances. We present an algorithm for induction of fuzzy conjunctive rules. This algorithm is based on the set covering paradigm that uses fuzzy instead of crisp sets to induce fuzzy classification rules. This paper investigates three research questions: (1) the effect of four novel evaluation functions adapted to the fuzzy set domain for this concept learning algorithm, (2) the search paths followed in a fuzzy lattice, and (3) the benchmark results for each evaluation function on nine data sets.