结合初始中心优化和特征加权的K-Means聚类算法

为了提高传统K-Means聚类算法的聚类准确性,提出一种结合初始中心优化和特征加权的改进K-Means聚类算法。首先,根据样本特征对聚类的贡献程度获得初始特征权重,构建一种加权距离度量。其次,利用提出的初始聚类中心选择方法获得k个初始聚类中心,并结合初始特征权重进行初步聚类。然后,根据聚类精度来调整特征权重并再次执行聚类过程。重复执行上述过程直到聚类精度不再变化,获得最终的聚类结果。在UCI数据库上的实验结果表明,与现有相关K-Means聚类算法相比,该算法具有较高的聚类准确性。

K-Means Clustering Algorithm Based on Initial Center Optimization and Feature Weighted

In order to improve the clustering accuracy of traditional K-Means clustering algorithm,an improved K-Means clustering algorithm based on initial center optimization and feature weighted was proposed.Firstly,the initial feature weight is obtained based on the contribution factor of sample feature for clustering,and a weighted distance metric is constructed.Next,the k initial clustering centers are obtained by using the proposed initial clustering center selection method,and the initial clustering is performed with the initial feature weight.Then,the feature weights are adjusted according to the clustering accuracy and the clustering process is performed again.The above process is repeated until the clustering accuracy is no longer changed,resulting in the final clustering result.The experimental results on the UCI database show that the algorithm has high clustering accuracy compared with the existing K-Means clustering algorithm.