一种改进的最小距离分类器NN-MDC

An Improved Method for Minimum Distance Classify:NN-MDC

为了提高最小距离分类器的性能,在其基础上提出了一种改进MDC——NN-MDC:它先对训练样本进行修剪,根据每个样本与其最近邻类标的异同决定其取舍,然后再用剩余的训练样本训练得到分类器。采用UCI标准数据集实验,结果表明本文所提出的NN-MDC与MDC相比具有较高的分类精度。

Minimum distance classifier is a simple and effective classification method. It is the statistical average of various samples as the basic template, then calculate the distance from the base template to test sample. However, the classification performance is very bad, when the training sample is not clustering. To improve its classification performance, the improved method is proposed,which named NN- MDC. First, the training sample is pruning, then classifier is trained by remaining training samples that. Experiment is tested on UCI standard database,the experimental results show that the proposed nearest neighbor minimum distance classifier is effective,and it is superior to minimum distance classifier in classification performance.