Abstract: In order to improve the precision and diversity of individual networks to improve the generalization perfor-mance of neural network ensemble (NNE),this paper proposed a method to generate individual neural network for ensembling based on twice clustering.Firstly,all the samples are chosen to cluster for the first time to form once clustering subclass,and then the twice clustering is performed for each type of subclass to form the sample subsets of each subclass.Affinity Propagation(AP) clustering makes the otherness criterion of “similar in classes,diversity between classes” maxization,and the samples in the class can response the real data distribution.Finally,according to the permutation and combination a subset is selected from each twice clustering of each subclass to construct a trainset to train an individual neural network.So the trainset with the smaller size of the data and the real data distribution can train the individual neural networks with the bigger diversity,and the ensemble of these individual neural networks can get better performance.Simulation experiments show that our proposed method here is effective.

Key words: Affinity propagation(AP) clustering,Neural network ensemble(NNE),Twice clustering,Individual neural network