Simultaneous Concept Learning of Fuzzy Rules

FuzzyBexa was the first algorithm to use a set covering approach for induction of fuzzy classification rules. It followed an iterated concept learning strategy, where rules are induced for each concept in turn. We present a new algorithm to allow also simultaneous concept learning and the induction of ordered fuzzy rule sets. When a proper rule evaluation function is used, simultaneous concept learning performs far better than iterated concept learning with respect to rule set size, rule complexity, search complexity, and classification accuracy. We provide empirical results of five experiments on nine data sets and also show that the algorithm compares favourably to other well known concept learners.