Improving a Fuzzy ANN Model Using Correlation Coefficients

An improvement of a fuzzy Artificial Neural Network model based on correlation coefficients is presented. As aggregation operator to compute the net input to target neuron a uninorm is used, instead of the sum of all the influences that the neuron receives usually used in typical artificial neurons. Such combination allows increasing the model performance in problem solving. While the natural framework and the interpretability presented in the former model are preserved by using fuzzy sets, experimental results show the improvement can be accomplished by using the proposed model. Significant differences of performance with the previous model in favor of the new one, and comparable results with a traditional classifier were statistically demonstrated. It is also remarkable that the model proposed shows a better behavior in presence of irrelevant attributes than the rest of tested classifiers.