Adaptive mean shift clustering based on cover-tree
温柳英 1庞柯1
扫码查看
点击上方二维码区域，可以放大扫码查看
作者信息
1. 西南石油大学计算机科学学院,四川成都 610500
折叠
摘要
为解决均值漂移聚类算法聚类效果依赖于带宽参数的主观选取,以及处理密度变化大的数据集时聚类结果精确度问题,提出一种基于覆盖树的自适应均值漂移聚类算法MSCT(MeanShift based on Cover-Tree).构建一个覆盖树数据集,在计算漂移向量过程中结合覆盖树数据集获得新的漂移向量结果KnnShift,在不同数据密度分布的数据集上都能自适应产生带宽参数,所有数据点完成漂移过程后获得聚类结果.实验结果表明,MSCT算法的聚类效果整体上优于MS、DB-SCAN等算法.
Abstract
To solve the problem that the clustering effects of the mean-shift clustering algorithm depend on the subjective selec-tion of bandwidth parameters and the problem of the accuracy of the clustering results when dealing with datasets with large den-sity changes,an adaptive mean-shift clustering algorithm based on covering tree-MSCT was proposed.A cover tree data set was built,and the cover tree data set was combined to obtain a new drift vector result KnnShift in the process of calculating the drift vector,which adaptively generated bandwidth parameters on data sets with different data density distributions,and the cluste-ring results were obtained after all data points completed the drift process.Experimental results show that the clustering effect of MSCT algorithm is better than that of MS and DBSCAN.
维普
万方数据