A novel nonparametric clustering algorithm for discovering arbitrary shaped clusters
Yu He, Lihui Chen · 2004
Most existing clustering algorithms have at least one of the two following problems. They either require users to carefully set or tune some predefined parameters, or they have difficulty in discovering arbitrary shaped clusters. To solve these two problems, a nonparametric clustering algorithm called MinClue (MINimum spanning tree based CLUstEring) aiming at discovering arbitrary shaped clusters is proposed in this paper. It first constructs a minimum spanning tree (MST) of the dataset and then automatically decides a threshold for removing inconsistent edges from the MST. The experimental result demonstrates the effectiveness of MinClue.