Saw-Tooth Algorithm Guided by the Variance of Best Individual Distributions for Designing Evolutionary Neural Networks

摘要

This paper proposes a diversity generating mechanism for an evolutionary algorithm that determines the basic structure of Multilayer Perceptron (MLP) classifiers and simultaneously estimates the coefficients of the models. We apply a modified version of a recently proposed diversity enhancement mechanism [1], that uses a variable population size and periodic partial reinitializations of the population in the form of a saw-tooth function. Our improvement on this standard scheme consists of guiding saw-tooth reinitializations by considering the variance of the best individuals in the population, performing the population restart when the difference of variance between two consecutive generations is lower than a percentage of the previous variance. The empirical results over six benchmark datasets show that the proposed mechanism outperforms the standard saw-tooth algorithm. Moreover, results are very promising in terms of classification accuracy, yielding a state-of-the-art performance.