Training Artificial Neural Networks Using Lévy Group Search Optimizer
Inspired by animal behavior, especially animal searching behavior, a novel swarm intelligence algorithm Group Search Optimizer (GSO) has been proposed recently . In this paper, we propose a new artificial neural network (ANN) training algorithm based on an improve GSO algorithm. We replace the gaussian random walk in the standard GSO with Lévy flight, which is a random search patterns adopted by many organisms to maximize the efficiency of resource searches in uncertain environments. We firstly evaluate the improved GSO with Lévy flight (LGSO) on a set of 5 optimization benchmark functions. We then apply the LGSO algorithm to tune the parameters of a 3-layer feed-forward ANN, including connection weights and bias. Two real-world problems, namely Cleveland heart disease classification problem and sunspot number forecasting problem, have been employed to assess the performance of our LGSO-trained ANN (LGSOANN). In comparison with other sophisticated machine learning techniques proposed in recent years such as ANN ensembles, LGSOANN has better convergence and generalization performance on the two real-world problems.
Keywords: Optimization, Animal Behavior, Group Search Optimizer, Lévy Flight, Swarm Intelligence, Evolutionary Algorithm.