Revisiting the Wang–Mendel algorithm for fuzzy classification

Abstract

[Abstract]: In this paper, we review the Wang–Mendel algorithm for the induction of fuzzy IF-THEN rules in the context of classification problems. A general fuzzy inference architecture for classification is proposed with the aim of studying the influence of alternative configurations of the learning model. Specifically, we analyse the effects of changing the aggregation strategy, and we explore the use of different rule definitions, including or not the possibility to assign weighting factors to the generated rules. We test different rule weighting heuristics at this respect. The notion of rule conflict introduced in earlier versions of the algorithm is also reviewed in the context of the various resulting configurations of the fuzzy inference engine. A generalized version of the algorithm therefore results, bringing more flexibility to the configuration of the fuzzy inference engine, and improving the performance for certain problems. The main objective is to complement the results of previous approaches by offering a comprehensive overview of this popular algorithm for fuzzy rule induction in the context of classification problems. Several well-known machine learning classification benchmarks are analysed and compared looking for the best possible model configuration.