A dynamic method that emphasizes diversity for constructing ensembles of neural network classifiers

It is well known that ensembles of neural network classifiers produce better accuracy than a single neural classifier provided there is diversity in the ensemble. In this paper we present a dynamic method for producing such ensembles that emphasizes diversity in the ensemble members by weighted k-nearest neighbors. This emphasis on diversity produces ensembles with low generalization errors from ensemble members with comparatively high generalization error. We compare this with other methods on performance, and find that our method is efficient and effective.