The back-propagation algorithm is generally used to train a Feed-forward Neural Network (FNN) on a dataset to learn the prediction rules. However, the incremental adaptation approach of the back-propagation (BP) algorithm is known to take long computation times for training. Besides, it is also found to be susceptible to the local minima. Moreover, attempts to improve the performance of the BP algorithm by combining it with other algorithms are inadequate. In this study, we apply the Particle Swarm Optimization (PSO) to train an FNN. PSO is a recently invented evolutionary computation technique which is gaining popularity owing to its simplicity in implementation and rapid convergence. The BP method and the PSO method for training FNNs on three well-known models are presented. The experimental results demonstrate that the simple PSO algorithm without any special modification to serve as a neural training algorithm produces better results than the slow and time-consuming BP algorithm.

optimization, neural networks, machine learning, particle swarm optimization.