Classification using neural networks trained by swarm intelligence

The metaheuristics are the algorithms that are designed to solve many optimization problems without needing knowledge about the corresponding problems in detail. Similar to other metaheuristics, the Migrating Birds Optimization (MBO) algorithm which is introduced recently is a nature inspired neighbourhood search method. It simulates migrating birds’ V flight formation which is an effective flight shape for them to save the energy. In this paper, 20 different data sets were used for classification. Firstly, the MBO algorithm was employed to train neural networks which were designed for classification. Then, the same networks were trained by using other well-known powerful metaheuristic algorithms. These are the Artificial Bee Colony (ABC) algorithm, the Particle Swarm Optimization (PSO) algorithm, the Differential Evolution (DE) algorithm and the Genetic Algorithm (GA). Finally, the Levenberq-Marquardt (LM) algorithm, a classical gradient based training method, was added to implementations so that clear comparisons could be done among algorithm performances. Results show that the MBO algorithm has better performance than the others’ performances. It gets the highest accuracies in tests and reaches to the lowest mean squared errors in trainings for most of the experiments.