Optimal Rule Extraction of RBFN Based System Using Hierarchical Self Organised Evolution

Abstract The fuzzy if-then rule extraction invariably assumes a preassigned structure instead of an optimal one. The paper presents the development of a hierarchical self organized radial basis function network (RBFN) that simultaneously evolve the structure and parameter of the Fuzzy rule-base. Robust particle swarm optimization (RPSO) is used as a tool for the learning of the state reproducing the result repeatedly with a preassigned value of iteration. Also the multi dimensional crossover vector is introduced as a set of Accommodation Boundary of the data set to employ desired number of linguistic fuzzy rules. Experiments conducted and comprehensive analyses show that the proposed method produces smaller number of rules with respect to the other methods along with comparable error. Also the computational time for learning will decrease significantly in this method as the concept of iteration during a learning cycle has been removed. The effect of different membership function has also been studied during the recruitment of node.