Data base definition and feature selection for the genetic generation of fuzzy rule bases

摘要

This paper presents further investigations with a fuzzy genetic method for the generation of fuzzy rule bases using improved preprocessing conditions. This method, previously proposed by the authors, uses the degree of coverage of candidate rules to select the ones forming the search space of the genetic algorithm, thus it is named DOC-BASED method. For this particular fuzzy genetic method, a previous attribute selection might be necessary. The definition of the fuzzy data base can also impact the results of the automatic generation of rule bases. To this end, an heuristic method for database definition, FUZZY-DBD and an attribute selection method, FUZZY-WRAPPER, were previously proposed and investigated in different contexts by the authors. The goal of this paper is to propose and investigate an enhanced version of the DOC-BASED method, regarding time reduction for the genetic search space generation, combined with the FUZZY-DBD and the FUZZY-WRAPPER methods for the database definition and attribute selection process. The methods used are described and the advantages of the enhanced version of the DOC-BASED method is discussed. Experiments were performed aiming at comparing results generated by the original plain DOC-BASED method and its extended version described here. The experiments also include studies using a filter-based attribute selection method. Experimental results on 10 datasets are presented and compared. Results show that a fuzzy approach to attribute selection and their proper fuzzification can yield significant improvements to rules generation.