An Efficient Hierarchical Clustering Algorithm via Root Searching

As an important branch of machine learning, clustering is wildly used for data analysis in various domains. Hierarchical clustering algorithm, one of the traditional clustering algorithms, has excellent stability yet relatively poor time complexity. In this paper, we proposed an efficient hierarchical clustering algorithm by searching given nodes´ nearest neighbors iteratively, which depends on an assumption: the representative node (root) may exist in the densest data area. The experiments results preformed on 14 UCI datasets show that our algorithm exhibits the best accuracies on most datasets. Moreover, our method has a linear time complexity which is significantly better than other traditional clustering methods like UPGMA and K-Means.