An Efficient Approach to Optimize the Learning Rate of Radial Basis Function Neural Network for Prediction of Metastatic Carcinoma

Abstract

This paper is based on the optimization of linear weights in a radial basis function neural network that connects the hidden layer and the output layer. A new optimization algorithm called dualist algorithm is applied for choosing an optimal learning parameter. A conventional strategy of random selection of radial basis function (RBF) centers and the width, as well as the weights, are estimated by gradient descent method and least square methods, respectively. The ideology behind this study is to predict the occurrence of metastatic carcinoma in human cells by computational approaches. Our simulation consists of comparing the predictive accuracy of harmony search-radial basis function network (RBFN) and dualist-RBFN by optimizing the weight factor. The Wisconsin breast cancer dataset is used as a benchmark to experiment our training pattern. The learning rate (weight factor) is taken as the optimized parameter to obtain the best possible solution.