A neuro-fuzzy approach to aerobic fitness classification: a multistructure solution to the context-sensitive feature selection problem

"Divide and conquer" is an old problem-solving principle. In this paper, we propose a neuro-fuzzy approach to the problem of context-sensitive feature selection. The method is based on a fuzzy preclassifier and multistructure feedforward neural networks. The same feature sets are used in the different structures, but the neurocalculation is different in each structure due to different synaptic weights. Finally, the classification output is selected from the multi-network structure using the fuzzy preclassifier's output as a decision criterion. The proposed method is tested in a case of human fitness classification. The research material was clinically obtained and the test subjects were healthy adults. The results are compared to a those obtained with a conventional neural network classifier. The method proposed here using a neuro-fuzzy approach to classification improved the classification accuracy by over 10 percent.