Ensemble approach for improving generalization ability of neural networks

This paper presents a study on improving generalization ability of neural networks (NNs) by using ensemble approach. In already existing literature, both theoretical and experimental studies have revealed that the performance, i.e., generalization ability of NN ensemble is greatly dependent on both accuracy and diversity among individual NNs in the ensemble. In this study and implementation of NN ensemble, Back Propagation (BP) learning algorithm is used to train individual NNs independently for a fixed number of training epoches. We have considered 12 different benchmark problems in our study. Few papers have considered such a large number of problems. The experimental results show that the performance of NN ensemble is often better than individual NNs, and both accuracy and diversity among participating networks are important for the generalization ability of the ensemble.