A Fuzzy Complementary Criterion for structure learning of a neuro-fuzzy classifier

In this paper, the use of a Fuzzy Complementary Criterion (FuzCoC) for structure learning of a neuro-fuzzy classifier arranged in layers is proposed. The FuzCoC has been recently proposed as an effective criterion for feature selection. Simulation results in a large number of benchmark problems revealed the capability of this method in selecting small subsets of powerful and complementary features even in high dimensional feature sets. In this paper, the FuzCoC method is used not only to reduce the dimensions of the original feature space, but also to identify complementary generic fuzzy neuron classifiers (FNCs) arranged in layers. The chosen generic classifiers are then combined using a decision fusion operator to construct a descendant FNC at the next layer with enhanced classification capabilities. The proposed structure learning algorithm is a modified version of the Group Method of Data Handling (GMDH) algorithm which incorporates the FuzCoC method simultaneous as a pre-feature selection method and as a method to identify complementary generic classifiers to be combined in the next layer. Simulation results demonstrate the capabilities of the proposed method in building accurate neuro-fuzzy classifiers with reduced computational demands.