An Effective Clustering Algorithm for Auto-Detecting Well-Separated Clusters
Jinyuan He, Gansen Zhao, Haolan Zhang, Kotagiri Ramamohanarao, Chaoyi Pang · 2014
Clustering is an important analysis method commonly used in many areas, including data mining, image processing, statistics, biology, and machine learning. In this paper, we introduce a novel effective clustering method based on Euclidean Distance called Self-Increase Clustering (SIC) for detecting well-separated clusters that can be either convex or non convex sets. Unlike most of the prevalent clustering algorithms, SIC does not require any initial parameters such as the number of clusters produced. Instead, SIC can discover the clusters number automatically based on the distribution of input data and separate these clusters effectively. In each iteration of SIC, a new cluster containing one randomly selected object is created and then this cluster increases by merging itself with the other objects or clusters near-by if certain criterion is satisfied. We evaluate SIC both from theoretical as well as practical points of view, and the experimental results show that SIC works effectively and efficiently on different data sets.