Ensemble of duo output neural networks for binary classification

This paper presents an ensemble of duo output neural networks (DONN) using bagging technique to solve binary classification problems. DONN is a neural network that is trained to predict a pair of complementary outputs which are the truth and falsity values. Each component in an ensemble contains two DONNs in which the first network is trained to predict the truth and falsity outputs whereas the second network is trained to predict the falsity and truth outputs which are set in reverse order of the first one. In this paper, we propose classification techniques based on outputs obtained from DONNs. Also, the ensemble selection technique is proposed. This technique is created based on uncertainty and diversity values. All proposed techniques have been tested with three benchmarks UCI data sets, which are ionosphere, pima, and liver. It is found that the proposed ensemble techniques provide better results than those obtained from an ensemble of back propagation neural networks, an ensemble of complementary neural networks, a single pair of duo output neural networks, a single pair of complementary neural networks, and a back propagation neural network.