A Greedy Approach to Rule Reduction in Fuzzy Models

The characteristics of a fuzzy model are frequently determined by the manner in which the rules are constructed. Rules obtained by a heuristic assessment of a system generally are linguistically interpretable and have large granularity. The generation of rules via learning algorithms that analyse training data produces precise models consisting of multiple rules of small grannularity. In this paper, a greedy algorithm is presented that combines rule learning with a region merging strategy to reduce the number of rules. This approach differs from standard rule reduction techniques in that the latter are employed after the rule base has been completed while the learn-and-merge strategy generates a rule simultaneously with expanding its region of applicability. The objective of the algorithm is to produce fuzzy models with both a small number of interpretable rules and high precision.