优化初始聚类中心的改进k-means算法

Improved k-means algorithm with meliorated initial centers

传统k-means算法随机选取初始聚类中心使聚类结果不稳定,诸多优化算法的时间复杂度较高,为了提高聚类稳定性并降低时间复杂度,提出了基于个体轮廓系数自适应地选取优秀样本以确定初始聚类中心的改进k-means算法。该算法多次调用传统k-means算法聚类,根据k个类中心的个体轮廓系数以及各样本与类中心的距离,自适应地选取优秀样本,求其均值作为初始聚类中心。在多个UCI数据集上的实验表明,该算法聚类时间短,具有较高的轮廓系数和准确率。

The initial centers of the traditional k-means algorithm are randomly selected, so that the clustering results are uncertain. Many improved algorithms have a high time complexity. In order to enhance stability of clustering and reduce the time complexity, improved k-means algorithm which selects good samples adaptively based on individual silhouette coefficient to get initial centers is presented. Traditional k-means algorithm is used for some times to cluster data according to, the individual silhouette coefficient of k class centers and the distance between samples and class centers, good samples are choosen adaptively, the average value is used as initial centers. Tests on some UCI data sets show that this algorithm has shorter time to cluster data, higher silhouette coefficient and higher accuracy.