Improving The Design Of Fuzzy Classifiers Using Multi Objective Evolutionary Algorithms

Abstract

The development of knowledge based system is very complex due to its newlinesubjective nature of human knowledge representation and manipulation in newlineknowledge based systems. This leads to the implementation of many human newlinedecisions making into machine like interment of new information, common newlinelogic/sense, capacity to reason, adaptations to unknown situations etc. newlineTo deal with the uncertainty and imprecision inherent in human knowledge, newlinefuzzy logic has been used due to its strong mathematical framework. These newlinesystems are known as Fuzzy Knowledge Based Systems (FKBS) , newlinealternatively known as Fuzzy Rule Based Systems (FRBS) . newlineDevelopment of FKBS leads to optimization of the performance and it may be newlineseen as optimization task. For this purpose genetic algorithms are used to newlinerepresent the different parameters of FKBS. These systems are known as newlineGenetic Fuzzy Systems (GFS). newlineFurther multi-objective optimization is also used to design and implement newlinefuzzy systems with the usage of multi-objective evolutionary algorithms newline(MOEA). newlineDuring the design of FKBS, there are two important parameters named newlineinterpretability and accuracy which are used to measure the performance of newlineFKBS. Interpretability is the subjective feature of the FKBS that shows how newlinemuch the functioning of FKBS is understandable by its user. However, newlineaccuracy shows the closeness between the real and modeled systems. The newlinemulti-objective formulation of FKBS includes the objectives; interpretability newlineand accuracy. The overall goal is to design and implement interpretable newlineFKBS with higher accuracy. newlineThe research is carried out into two parts. First section deals with the newlinecreation/formulation of a fuzzy knowledge base system. In this section, a new newlineFKBS has been proposed with Hierarchal Fuzzy Partitioning (HFP) and two newlinedifferent rule generation methods Wang and Mendel and Fuzzy Decision Tree. newlineThe initial FKBS is developed using type-1 fuzzy theory. Second section is newlinethe multi-objective formulation of FKBS. This formulation is implemented newlineusing five different MOEAs. In this form