An Evolutionary Approach for Optimizing Three-Layer Perceptrons Architecture

We propose an evolutionary algorithm for optimizing the hidden layer size of three-layer perceptrons. The optimization problem is posed in terms of finding, for each learning database, the best number of neurons to use in the hidden layer. For this, a population of three-layer perceptrons is evolved using the mean squared error as a measure of fitness. Each individual of this population is trained using the backpropagation learning algorithm. During the evolutionary process, parents are chosen using the rank selection operator and new candidate solutions are produced using the two-point crossover and mutation operators. Experiment results show that the proposed method perform well for different examples of real test data. Typical examples of these results are presented and discussed.