基于二路生成树的聚类边界检测算法

BOUNDARY DETECTING ALGORITHM FOR CLUSTERING BASED ON TWO ROUNDS OF MINIMUM SPANNING TREES

聚类的边界是一种有用的模式,为有效地提取聚类的边界点,提出c-层近邻概念,将c-层近邻应用于二路生成树,能快速计算出每个对象的反向近邻值,从而根据反向近邻值提取聚类的边界。提出的基于二路生成树的边界检测算法(DBMST)在综合数据集和真实数据集的实验结果表明,该算法在含有噪声/孤立点的数据集上,能够快速有效地识别出聚类的边界。

Clustering boundary is a useful pattern. In order to extract the boundary points of cluster effectively, we propose a c-layer nearest neighbour concept, and apply it to two rounds of minimum spanning trees. This novel conception can count every point＇ s reverse nearest neighbouring value rapidly, thus according to the value the boundary of cluster is extracted. Results of the experiment of proposed boundary detecting algorithm based on two round minimum spanning carried out on both synthesise datasets and real datasets demonstrate that the algorithm can fast identify the boundary of clusters effectively on the datasets with noise/outlier points.