Intra-specific competitive co-evolutionary artificial neural network for data classification

Abstract

This paper presents an alternative approach of competitive co-evolutionary (ComCoE) artificial neural network (ANN) developed for data classification. The motivation of this work is to employ an interactive game-based fitness evaluation method within a CoE framework to develop a compact and accurate ANN model. The proposed model uses only one population of radial basis function artificial neural networks (RBFANNs) in the CoE framework to find out an optimised RBFANN. In the ComCoE process, the RBFANNs compete in an intra-specific competition environment, which is driven by a game-based fitness evaluation method. The fitness evaluation for each RBFANN is made by computing the interaction among the selected RBFANNs in a population quantitatively throughout a number of encounters under a Single Elimination Tournament topology. Two indicators, i.e. the classification accuracy and hidden nodes number of each RBFANN, are referred to compute the fitness value. The proposed model performs a global search for finding potential near optimal solution. Then, a local search (Backpropagation algorithm) is executed to reach at a precise solution. The proposed classification model is evaluated using 14 public data sets from the UCI machine-learning repository. A performance comparison between the proposed model and other state-of-art classifiers is also conducted. The empirical results show that the proposed model, which constructs a compact network structure, could perform with high classification accuracy rates.