Learning enhancement of radial basis function neural network with harmony search algorithm

Abstract

Training Radial Basis Function (RBF) neural network with Particle Swarm Optimization (PSO) was considered as a major breakthrough, that overcome the stuck to the local minimum of Back Propagation (BP) and time consuming and computation expensive problems of Genetic Algorithm (GA). However, PSO proved some problems to achieve the goal, i.e., it converged too fast so that it stuck to the local optimum. Furthermore, particles may move to an invisible region. Therefore, to realize the enhancement of the learning process of RBF and overcome these PSO problems, Harmony Search Meta-Heuristic Algorithm (HSA) was employed to optimize the RBF network and attain the desired objectives. The study conducted a comparative experiments between the integrated HSA-RBF network and the PSORBF network. The results proved that HSA increased the learning capability of RBF neural network in terms of accuracy and correct classification percentage, error convergence rate, and less time consumption with less mean squared error (MSE). The new HSA-RBF model provided higher performance in most cases and promising results with better classification proficiency compared with that of PSO-RBF network.