A Novel Neural Network Ensemble Method Based on Affinity Propagation Clustering and Lagrange Multiplier
Helong Yu, Guifen Chen, Dayou Liu, Baocheng Wan, Di Jin · 2009
To improve the forecasting precision and generalization capability of neural network, a novel neural network ensemble method is proposed, in which bagging algorithm is used to generate neural network individuals and root of mean square error is adopted as a rule to measure the similarity between networks.By the affinity propagation clustering algorithm, neural network individuals with high precision and strong diversity are selected. Then by the Lagrange multiplier method, these optimally selected neural networks are combined. The test on the standard dataset shows that the ensemble method proposed in the paper is of higher forecasting precision and better generalization capability than the single network and the neural network ensemble method based on forecasting effective measure method.