基于密度峰值剪枝后的最短路径聚类算法

Clustering by Pruning Paths Based on Shortest Paths from Density Peaks

聚类是通过数据标签或者属性,将一系列经验数据按照相似性或者相近性进行归类.基于密度属性展开的聚类算法,主要聚焦在聚类中心的确定和剩余点如何分配的问题上展开讨论.针对基于密度峰值的可训练最短路径算法,通过密度峰值确定聚类中心,提出使用截断阈值、对路径图进行剪枝的算法改进.然后基于最短路径法对剩余点进行全局分配.实验结果证明,在保持聚类精度的同时,有效地提升了算法执行效率.

Clustering is to classify multiple empirical data according to their similarity or proximity based on data labels and properties.For the clustering algorithm based on the density peaks,it mainly focuses on the determination of the clustering center and how to allocate the remaining points.In this paper,according to a trainable clustering algorithm based on shortest paths to density peaks,the clustering center is determined by the density peaks.We propose that using a cutoff threshold and pruning the path graph to improve the algorithm.The remaining points are allocated globally based on the shortest path method.It is proved that the algorithm can significantly improve the efficiency while maintaining the clustering accuracy.