Hybrid fuzzy classifier based on feature-wise membership given by artificial neural network

This paper proposes the hybrid neuro-fuzzy classification model to perform the supervised classification of the data. In the proposed classification model, artificial neural network is used to learn the membership function for fuzzy classes of an input data set. This learned membership function gives the belongingness of each feature value to all the classes. Then the feature selection based on the information gain is performed to select the subset of features that gives the highest classification accuracy. Only the feature belongingness of the selected features is aggregated by using sum aggregation reasoning rule to compute the pattern belongingness to the given classes. These values of feature or pattern belongingness are useful for high level decision making. Finally by using defuzzification operation each pattern is assigned to the predicted class. In this paper, the proposed model is applied to four datasets: IRIS, BUPA, PIMA and WINE. Accuracy of the results is measured by using the performance measures-Misclassification, Percentage of overall class accuracy and Kappa index of agreement. In addition to the improved performance, proposed model gives the freedom to handle each feature independently based on its belongingness value.