An Optimized and Efficient Radial Basis Neural Network using Cluster Validity Index for Diabetes Classification

This Radial Basis Function Neural Networks (RBFNNs) have been used for classification in medical sciences,especially in diabetes classification. These are three layer feed forward neural network with input layer, hidden layer and output layer respectively. As the number of the training patterns increases the number of neurons in the hidden layer of RBFNNs increases, simultaneously network complexity increases and classification time increases. Although various efforts have been made to address this issue by using different clustering algorithms like k-means, k-medoids, and Self Organizing Feature Map (SOFM) etc. to cluster the input data of diabetic to reduce the size of the hidden layer. Though the main difficulty of determination of the optimal number of neurons in the hidden layer remains unsolved. In this paper, we present an efficient method for predicting diabetics using RBFNN with optimal number of neurons in the hidden layer. This study mainly focuses on determining the number of neurons in hidden layer using cluster validity indexes and also find out the weights between output layer and a hidden layer by using genetic algorithm. The proposed model was used to solve the problem of detection of Pima Indian Diabetes and gave an accuracy of 73.50%, which was better than most of the commonly known algorithms in the literature. And also proposed methodology reduced the complexity of the network by 90% in terms of number of connections,furthermore reduced the classification time of new patterns.