A Novel Method to Optimize the Structure of BP Neural Networks

Abstract

It has been a long time that there is not a so good method to determine the number of neurons in hidden layer for BP neural network. For this problem, a novel algorithm based on Akaike Information Criterion (AIC) to optimize the structure of the BP neuron networks is proposed in this paper. At the same time, this paper gives the upper and lower bounds for classical AIC to overcome its shortcomings. The simulation experiment shows that this method can select a more suitable network structure, and can ensure the minimal output error with the optimal structure of the network. DOI: http://dx.doi.org/10.11591/telkomnika.v11i10.3344