An evolutionary method to generate Mamdani rule-based systems with hedge algebra based semantics for regression problems

Abstract

In this paper, we propose an evolution algorithm to generate Mamdani Fuzzy Rule-based Systems (MFRBS) with different trade-off between complexity and accuracy. The algorithm was developed taking the idea of the schema evolution (2+2)M-PAES which has been proposed in [6]. The main novelty of the algorithm is to learn concurrently rule bases, fuzzy partitions and linguistic terms along with their fuzzy sets using hedge algebra (HA) methodology. The algorithm allows to generate rules from pattern data utilizing new information of partitions and fuzzy sets in the same individual. In addition, we propose a new method for encoding individuals that can be realized in the hedge algebra approach to solve this problem. The computer simulation is carried out with six standard regression problems in [10] accepted by the research community and the obtained results show that the MFRBSs generated by the proposed algorithm are better than those examined in [8] with respect to two objectives, the complexity and the accuracy.