Research On Parameter Optimization Based On Fuzzy Extension Matrix

Nowadays extension matrix learning has become an important branch of inductive learning. Due to introduce of the fuzzy idea, the fuzzy extension matrix can deal with uncertainty associated with human thinking and perception, so it is used more and more widely. During the process of the rule extraction, the cases covered by the rule have some overlap, so the entire process of building fuzzy extension matrix is based on α,β,γ. The introduce of parameters can reduce such overlap to some extent, decrease the uncertainty of classification and improve the accuracy of classification result. Thus, how to select the three important parameters is the central problem of the paper. In this paper, we firstly discuss the theory of extension matrix and the fuzzy extension matrix; the decision tree and the extension matrix are compared as well as the extension matrix and the fuzzy extension matrix; we propose the idea of optimizing fuzzy extension matrix parameters by the genetic algorithm based on this theory. Secondly, we further analyzed the sensibility of parameters to the fuzzy extension matrix classification result on the training accuracy, the testing accuracy and the rule number. And the implementing scheme of the optimizing parameters method with genetic algorithm is discussed in detail. Finally, this method has been verified through experiment based on Borland C++ Builder development platform and Matlab. The experimental results show that the optimal parameter value gained by this method makes fuzzy extension matrix obtaining the best classification result. So, getting optimization parameter by GA is a good way to gain the best classification result based fuzzy extension matrix.