A hierarchical clustering algorithm based on density for data stratification
Zhiwei Sun · 2012
Cluster analysis is a primary method for data mining. Existing cluster approaches use global input parameters. But the real data can't be described by them, and each input parameter will have a significant effect to the result. A new algorithm named HDBSCAN will be introduced for the purpose of cluster analysis which can produce nature stratification. The algorithm have a preprocess procedure which use graph to express the structure of neighborhood, thus the parameters can be easily set, this is important for determination of input parameters; and then a hierarchical approach based on density clustering algorithm is used to analysis the data with the different Eps-neighborhood. At last the relationship among the cluster results will be got by scanning the cluster results above. We show how to get the intrinsic clustering structure and show the results. Both theory analysis and experimental results confirm the approach can cluster data with automatic setting different parameters in different partitions.