Simultaneous evolution of neural network topologies and weights for classification and regression

Abstract

Artificial Neural Networks (ANNs) are important Data Mining (DM) techniques. Yet, the search for the optimal ANN is a challenging task: the architecture should learn the input-output mapping without overfitting the data and training algorithms tend to get trapped into local minima. Under this scenario, the use of Evolutionary Computation (EC) is a promising alternative for ANN design and training. Moreover, since EC methods keep a pool of solutions, an ensemble can be build by combining the best ANNs. This work presents a novel algorithm for the optimization of ANNs, using a direct representation, a structural mutation operator and Lamarckian evolution. Sixteen real-world classification/regression tasks were used to test this strategy with single and ensemble based versions. Competitive results were achieved when compared with a heuristic model selection and other DM algorithms.