Training artificial neural network by krill-herd algorithm

Updating the weights of artificial neural networks (ANN) during training is one of the main challenges of model prediction by ANN. In recent years, there has been more attention to using meta-heuristic algorithms such as particle swarm optimization to resolve the infirmity of algorithm based on gradient. Accordingly, the krill-herd optimization algorithm for training ANN was suggested in this paper. In this method, three main components of the krill-herd optimization algorithm, i.e. motion induced by other krill individuals, foraging motion, and physical diffusion, had to update the weights of ANNs. Also, its performance was tested by training feed-forward artificial neural networks that could be used for classification. Result of extensive experiment on the datasets of UCI showed better performance of this method than previous approaches.