Clustering analysis is a very useful tool for discovering unknown knowledgeform dataset because of there is no needed pre-knowledge for training instead offinding group in the dataset. According to the categories form research, there are threemajor clustering methods: Hierarchical clustering algorithms, Partition clustering algorithms and density based clustering algorithms, but the characteristic that make density based clustering algorithm is more powerful to the unknown dataset than others is automatic detect cluster number and shape detect form cluster. Although there are many algorithms for different kinds of clustering problem, but none can handle all in once, however input parameter is the common drawback of algorithms. Without any understanding about the dataset, try-and-error is the only way find out the input parameter which has great influence to clustering result. In this thesis, we consider all point in the dataset to finding an optimal parameter for describing point density in its position. Considering density as a continuing force field, combine density with minimum spanning tree which saves neighbors information form high dimensional space to one dimensional space can easily find out clusters. As the experiment result show, this automatic method none needed any parameter from user but still works well.
