抄録
In this work, we have proposed an ensemble of weighted 1-NNbased on bi-modal perturbation of the training set. The 1-NNclassifier has higher variance so it is a suitable to construct anensemble. We have transformed the distances as weights so that thenearest instances are in effect in the predictions of the 1-NNclassifiers. In constructing the ensemble, we have taken a randomsubsample of the training set consisting a random subspace of thefeatures space. In this way, the instability is exerted on the base1-NN by taking a fraction of the full training instances and features.
