Segmented Hidden NN - An improved structure of feed forward NN

Abstract

There has been developed many method for the better convergence and generalization ability of neural network. Multilayer Perceptron (MLP) is made multi hidden layered structure for better performance. But in these types of structures still error from any output classes propagates in the backward direction which has a negative impact on the weight updating as well as overall performance because every output class is connected with every other hidden unit. In this paper an improved version of feed forward neural network structure has been proposed called Segmented Hidden Layer Neural Network (SHNN). In this proposed method the hidden layer is made segmented with respect to each output attributes so that the error form any output attribute only can influence the hidden nodes and weights which is connected with it. SHNN is extensively tested on seven real world benchmark classification problems such as heart disease, ionosphere, australian credit card, time series, wine, glass and soybean identification. The proposed SHNN outperforms the existing Backpropagation (BP) in terms of generalization ability and also convergence rate.