Training back-propagation neural network using hybrid fruit fly optimization algorithm

Back-propagation (BP) neural network is a multilayer feedforward neural network and has been widely used. However, back-propagation algorithm has some drawbacks, such as tendency to getting into partial minimal value and hidden layer nodes are determined by experience and slow convergence, in order to avoid this disadvantages, this work puts forward a hybrid of fruit fly optimization algorithm (FOA) and ant colony optimization algorithm (ACO) to resolve the aforementioned drawbacks. Firstly, the proposed hybrid method is applied to find best the initial pheromone distribution which includes heuristic factor, expectation factor and pheromone volatility. Secondly, ant colony algorithm is applied to optimize the network structure and parameters of back-propagation neural network to overcome drawbacks of BP. In this paper, the proposed algorithm is applied to Function optimization problems, UCI flowers classification and octane value prediction problem. The results are compared with ACO-based learning algorithm for back-propagation neural network and the resulting accuracy of BP trained with ACO, FOA and FOAACO. The experimental results show that FOAACO algorithm is much better than ACO and FOA for training BP neural network in terms of avoiding local optimum and converging speed. It is also proved that the proposed hybrid fruit fly optimization algorithm is better accuracy than ACO and FOA for training BP neural network. Copyright © 2016 American Scientific Publishers All rights reserved.