Neuro-fuzzy classifier based on the Gaussian membership function

This paper proposes the neuro-fuzzy classification model to perform the supervised classification of the data. In the proposed classification model, fuzzy membership matrix is formed by using Gaussian membership function. Membership matrix contains the membership of each feature value to the given classes. This membership matrix is given as an input to the artificial neural network and membership of each pattern to the given classes is obtained. Using the MAX defuzzification method, target class for each pattern is predicted. The proposed model is applied to four datasets: Iris, Pima, Bupa and Phoneme. The datasets were obtained from the University of California at Irvine (UCI) machine learning repository & ELENA database. Accuracy of the results for medical databases is measured by using the performance measures-Accuracy, Sensitivity & Specificity and that for non medical databases-Percentage of overall class accuracy and Kappa index of agreement. The performance of the proposed classifier is compared with the well known classifiers: Artificial neural network and C4.5 algorithm. The experimental results show that the proposed classifier gives the higher accuracy with good KIA values than these classifiers.