A Hybrid Gravitational Search Algorithm and Back-Propagation for Training Feedforward Neural Networks

Abstract

Presenting a satisfactory and efficient training algorithm for artificial neural networks (ANN) has been a challenging task. The Gravitational Search Algorithm (GSA) is a novel heuristic algorithm based on the law of gravity and mass interactions. Like most other heuristic algorithms, this algorithm has a good ability to search for the global optimum, but suffers from slow searching speed. On the contrary, the Back-Propagation (BP) algorithmcan achieve a faster convergent speed around the global optimum. In this study, a hybrid of GSA and BP is proposed to make use of the advantage of both the GSA and BP algorithms. The proposed hybrid algorithm is employed as a new training method for feedforward neural networks (FNNs). To investigate the performance of the proposed approach, two benchmark problems are used and the results are compared with those obtained from FNNs trained by original GSA and BP algorithms. The experimental results show that the proposed hybrid algorithm outperforms both GSA and BP in training FNNs.