Reckon the Parameter of DBSCAN for Multi-density Data Sets with Constraints
Tianqiang Huang, Yangqiang Yu, Kai Li, Wen-fu Zeng · 2009
DBSCAN is a typical density-based clustering algorithm, but it is time-consuming to ascertain the parameter Eps and it does not perform well on multi-density datasets because of the global parameter Eps. In this paper, we use must-link constraints to ascertain the parameter Eps for each density distribution effectively and automatically, which will be used to deal with multi-density data sets for traditional DBSCAN algorithm. Experimental results reveal that our algorithm can reckon the parameter of DBSCAN for multi-density data sets with constraints effectively.