Improved rule generation for a neuro-fuzzy network

The success of a neuro-fuzzy network is influenced by both its architecture and its learning algorithm. Currently, C.-J. Lin and C.-T. Lin's FALCON-ART algorithm ranks amongst the best structure/parameter learning algorithms yet devised. In this contribution, the FALCON-ART algorithm is adapted for use in neuro-fuzzy networks responsible for pattern recognition tasks. In contrast with FALCON-ART, each cluster is issued with its own vigilance parameter. Consequently, the sizes of individual rule antecedents can be controlled. A fuzzy logic controller is employed for this purpose. When it was applied to the Iris recognition problem, the neuro-fuzzy network attained an average recognition rate of 95.07 %. However, it fared slightly worse thana conventional neural network on a seismic signal discrimination task. The main advantages of the improved rule extraction algorithm are its speed, and the compactness of its resulting rule databases.