Título: Tests With Different Fitness Functions For Tuning Of Artificial Neural Networks With Genetic Algorithms
Autores: Lima, Aranildo R.; Mattos Neto, Paulo S. G. de; Silva, David A.; Ferreira, Tiago A. E.
Resumo: The choice of a good fitness function still a key element for the practitioners who use artificial intelligence to solve the forecasting problem. The fitness functions proposed in the literature have not been compared among them. Based on this fact, we started a brief empirical comparison among three different fitness functions in order to give some guidelines to help the fitness function choice. They were tested with a modified Genetic Algorithm for tuning the Artificial Neural Network structure and parameters. This experimental investigation with six non linear time series, showed that adjust the fitness function can be lead to a significantly improved accuracy for one given performance measure.
Palavras-chave: Time Series Forecasting; Genetic Algorithms; Fitness Function; Artificial Neural Networks; Artificial Intelligence
Código DOI: 10.21528/CBIC2011-32.5
Artigo em pdf: st_32.5.pdf
Arquivo BibTex: st_32.5.bib