| |
| Ensemble Learning to construct the learner in regression and classification is a hot research topic in machine learning in recent years, it has been proved empirically and theoretically to improve the generalization of learner in both regression and classification significantly, most of the methods to create Ensemble Learner by manipulating the training data, the architecture, the output of the neural network and others as evolutionary methods etc. Here, we present a new method to construct the ensemble learner through the behavior of the neural network-- the sensitivity of neural network that represents the nature of the neural network to some extent, and we give four measures based on the sensitivity of the neural network to select the neural network to represent the whole available neural network, some experiments based on the UCI benchmark prove that compare to the classic ensemble approaches as Bagging and Boosting , our method greatly improve the generalization of neural network ensembles and far smaller size both in regression and classification, the result of this paper gives us more suggestions that maybe we can pay more attention on the inhesion of neural network to construct the ensemble neural network with high diversity. |
|
| Keywords:Ensemble learning, sensitivity of neural network, select ensemble, machine learning, regression, classification |
|
