多尺度聚类挖掘算法

Multi-scale Clustering Mining Algorithm

数据挖掘领域在多尺度研究上已取得了一些进展。然而,当前研究主要集中于空间、图像数据的多尺度挖掘,并且传统的聚类挖掘并未对数据集的多尺度特性进行单独的研究。针对存在的问题,进行了普适性的多尺度聚类挖掘理论和方法的研究。首先,根据概念分层理论扩展尺度定义并构建多尺度数据集;其次,阐述尺度转换原因、分类,归纳多尺度聚类的定义;然后,以克里格法为理论基础,给出多尺度聚类尺度上推算法MSCSUA和多尺度聚类尺度下推算法MSCSDA;最后,利用公用UCI聚类数据集和H省全员人口真实数据集对算法进行实验验证,结果表明MSCSUA和MSCSDA是有效、可行的。

Data mining field has made some progress on the multi-scale research. However, the current research mostly focuses on the multi-scale mining of the space or image data. And traditional clustering mining has not separately stu- died the multi-scale characteristic of datasets. According to existing problems, this paper carried on the general study of multi-scale clustering mining theories and methods. Firstly, we extended scale definition on the basis of the concept hierar- chy and built multi-scale datasets. Secondly, we expounded the reasons and classification of scale conversion, meanwhile concluded the definition of the multi-scale clustering. Then, we introduced multi-scale clustering scaling up algorithm and multi-scale clustering scaling down algorithm based on the kriging theories. Finally, simulation experiments tested MSCSUA and MSCSDA with the help of public UCI clustering datasets and demographic dataset from H province. And the experimental results show that MSCSUA and MSCSDA are effective and feasible.