Improving Search Ability of Genetic Learning Process for High-Dimensional Fuzzy Classification Problems

Abstract

In this paper, we improve efficiency of the genetic search process for generating fuzzy classification rules from high-dimensional problems by using fitness sharing method. First, we define the similarity level of different fuzzy rules. It represents the structural difference of search space in the genetic population. Next, we use sharing method to balance the fitness of different rules and prevent the search process falling into local regions. Then, we combine the sharing method into a hybrid learning approach (i.e., the hybridization of Michigan and Pittsburgh) to obtain the appropriate combination of different rules. Finally, we examine the search ability of different genetic machine learning approaches on a suite of test problems and some well-known classification problems. Experimental results show that the fitness sharing method has higher search ability and it is able to obtained accurate fuzzy classification rules set.