基于双层结构的加速K-NN分类方法

Speeding K-NN classification method based on double-layer structure

在传统K-NN分类中,对于每个待测样本均需计算并寻找k个决策近邻,分类效率较低。针对该问题,提出一种双层结构的加速K-NN分类(K-NN classification based on double-layer structure,KNN_DL)方法。将正类和负类样本分别划分为多个不同子集,计算每个子集的中心和半径。当新样本进入时,选择k个决策近邻子集,若其具有相同的类别标签,将该样本标记为相应类别;反之,选择决策近邻子集中最近的k个决策近邻。这种双层结构的加速方式,压缩待测样本的决策近邻规模,提高效率。实验结果表明,KNN_DL方法能够获得较高的样本预测速度和较好的预测准确率。

For each sample to be tested of traditional K-NN classification,it is necessary to compute and select the k decision nearest neighbors,and the classification efficiency is low.To solve this problem,a speeding K-nearest neighbor(KNN)classification method based on double-layer structure(KNN_DL)was presented.The positive and negative samples were divided into a number of different subsets respectively,and the center and radius of all these subsets were computed.When a new sample was entered,the nearest k decision nearest neighbor subset was selected.If they had the same category label,the sample to be labeled was marked as the corresponding label.On the other hand,the k decision nearest neighbor of decision neighbor subset was selected.This double-layer speeding method compresses the size of the decision nearest neighbor set,and the learning efficiency is improved.Experimental results demonstrate that the KNN_DL model can obtain the high learning efficiency and testing accuracy simultaneously.