Hybrid Approach to Optimize the Centers of Radial Basis Function Neural Network Using Particle Swarm Optimization

Function approximation is an important type of supervised machine learning techniques, which aims to create a model for an unknown function to find a relationship between input and output data. The aim of the proposed approach is to develop and evaluate a function approximation models using Radial Basis Function Neural Networks (RBFN) and Particles Swarm Optimization (PSO) algorithm. We proposed Hybrid RBFN with PSO (HRBFN-PSO) approach, the proposed approach use PSO algorithm to optimize the RBFN parameters, depending to the evolutionary heuristic search process of PSO, here PSO use to optimize the best position of the RBFNN centers c, the weights w optimize using Singular Value Decomposition (SVD) algorithm and the Radii r optimize using K-Nearest Neighbors (Knn) algorithm, within the PSO iterative process, which means in each iterative process of PSO, the weights and Radii are updated depending the fitness (error) function. The experiments are conducted on three nonlinear benchmark mathematical functions. The results obtained on the training data clarify that HRBFN-PSO approach improved the approximation accuracy than other traditional approaches. Also, this result shows that HRBFN-PSO reduces the root mean square error and sum square error dramatically compared with other approaches.