EFuNN Ensembles Construction Using a Clustering Method and a Coevolutionary Multi-objective Genetic Algorithm

Abstract

AbstractThis paper presents the experiments which where made with the Clustering and Coevolution to Construct Neural Network Ensemble (CONE) approach on two classification problems and two time series prediction problems. This approach was used to create a particular type of Evolving Fuzzy Neural Network (EFuNN) ensemble and optimize its parameters using a Coevolutionary Multi-objective Genetic Algorithm. The results of the experiments reinforce some previous results which have shown that the approach is able to generate EFuNN ensembles with accuracy either better or equal to the accuracy of single EFuNNs generated without using coevolution. Besides, the execution time of CONE to generate EFuNN ensembles is lower than the execution time to produce single EFuNNs without coevolution.KeywordsExecution TimeRoot Mean Square ErrorTest PatternFuzzy Neural NetworkExecution Time AverageThese keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.