基于K-均值聚类的小样本集KNN分类算法

摘要

When KNN and its improved algorithms are performing classification, it always influences the final classification accuracy because of either too dense or too few the samples or too large the density differences among various kinds of samples. The paper proposes a small sample set KNN classification algorithm based on clustering technology. A new sample set is generated through clustering and editing which contains various kinds of samples with close densities. That new sample set is used to classify and label data objects whose classification and label numbers are unknown. Tests by standard data sets reveal that the algorithm can improve KNN classification accuracy and obtain satisfactory results.%KNN及其改进算法进行分类时,如样本集中、样本过少或各类样本的密度差异较大,都将会影响最后的分类精度.提出一种基于聚类技术的小样本集KNN分类算法.通过聚类和剪理,形成各类的样本密度接近的新的样本集,并利用该新样本集对类标号未知数据对象进行类别标识.通过使用标准数据集的测试,发现该算法能够提高KNN的分类精度,取得了较满意的结果.