Stabilized Nearest Neighbor Classifier And Its Statistical Properties

This paper is mainly studied?Stabilized Nearest Neighbor Classifier and Its Statistical Properties?,which is published by Will Wei Sun et al in 2016.The stability of statistical analysis is an important indicator for reproducibility,which is one main principle of scientific method.It means that similar statistical conclusions can be obtained based on two independent and identically distributed samples to predict new samples.However,in the classification context,most of Classifiers focuse on the precision without paying attention to the classification stability.In this paper,Will Wei Sun et al introduce a concept of classification instability(CIS)to quantify the sampling variability of the prediction made by a classification method.It is proved that the asymptotic CIS of any weighted nearest neighbor classifier turns out to be proportional to the Euclidean norm of its weight vector.Based on this concise form,the paper proposes a stable nearest neighbor classifier(SNN),which is obtained by controlling the accuracy of the weighted nearest neighbor classifier(WNN)in a small area and minimizing CIS.In theory,the paper proves that the convergence speed of CIS of SNN classifier has been optimized,and its improvement in CIS is greater than the loss in accuracy.Finally,through the simulation experiment and the simulation analysis of three sets of real data,it is verified that the SNN classifier has obvious improvement in CIS compared with the existing nearest neighbor classifiers such as kNN and BNN.