摘要
最小距离分类法和最近邻分类法是最简单、快速、有效的分类方法,但对噪声较敏感,对于训练样本很少或训练样本偏离类中心较远时,分类效果较差。针对这一问题,提出了基于均值更新(MU)的分类模型,通过不断扩大训练样本并更新均值中心来改善对测试数据的分类效果;并在此基础上提出了基于均值更新的最小距离(MU-MD)分类模型,利用MU的分类结果重新计算各类的均值,然后采用最小距离法对所有测试样本重新进行划分,以确定最终的类别归属,这样可以部分纠正MU分类过程中的错分,进一步提高分类效果。
The minimum distance classification algorithm and the nearest neighbor classification algorithm are the simplest, most rapid and most effective classification methods, and they are more sensitive to the noise. But to the training samples in few or the training samples that are far fi'om the cluster center, the classification results is poor. To solve this problem, this paper proposes a classification model based on the mean update (MU), by expanding the training sample and updating the mean center to improve the classification results of the test data; and on this basis, it proposes the MU-based minimum distance (MU-MD) classification model, and uses the MU's classification results to recalculate the mean of all test samples, then all test samples are re-divided by using the minimum distance method, so as to determine the final category attribution. This can partially correct misclassification in the MU category process and further improve the classification results.
出处
《计算机系统应用》
2012年第8期123-126,135,共5页
Computer Systems & Applications
关键词
最小距离分类法
均值更新
训练样本
测试样本
the minimum distance classification algorithm
mean update
training samples
test samples