Multi-Objective Hierarchical Genetic Algorithm for Modular Granular Neural Network Optimization

In this paper we propose a multi-objective hierarchical genetic algorithm (MOHGA) for modular neural network optimization. A granular approach is used due to the fact that the dataset is divided into granules or sub modules. The main objective of this method is to know the optimal number of sub modules or granules, but also allow the optimization of the number of hidden layers, number of neurons per hidden layer, error goal and learning algorithms per module. The proposed MOHGA is based on the Micro genetic algorithm and was tested for a pattern recognition application. Simulation results show that the proposed modular neural network approach offers advantages over existing neural network models. Finally the modular neural networks are joined using type-2 fuzzy integration, which allows having a system with a better behavior and results.