Development Of A Co-Evolutionary Radial Basis Function Neural Classifier By A K-Random Opponents Topology

The interest of the research in this paper is to introduce a novel competitive co-evolutionary (ComCoE) radial basis function artificial neural network (RBFANN) for data classification. The motivation is to derive a compact and accurate RBFANN by implementing an interactive "game-based" fitness evaluation within a ComCoE framework. In the CoE process, all individual RBFANNs interact with each other in an intra-specific competition. The fitness of each RBFANN is evaluated by measuring its interaction/encounter with k number of other randomly picked RBFANNs in the same population through a quantitative yet subjective manner under a k-random opponents topology. To calculate the fitness value, both the hidden nodes number and classification accuracy of each RBFANN are taken into consideration. To obtain a potential near optimal solution, the proposed model performs a global search through ComCoE approach and then performs a local search that is initiated by a scaled conjugate backpropagation algorithm to fine-tune the solution. Results from a benchmark study show high effectiveness of the co-evolved model with a k-random opponents topology in constructing an accurate yet compact network structure.