A Local Gravity Kinematics Synchronization Clustering Algorithm

Abstract

We propose a new clustering method called LGKSC, which is an alternative model based on gravitational kinematics to simulate local synchronization. The difference from existing clustering algorithms is that the algorithm makes objects over time. The dynamics of interaction are gradually synchronized dynamically, forming a local cluster corresponding to the internal structure of the data set. LGKSC can determine clusters with any shape, size and density, and can identify the amount of clusters automatically. LGKSC can adaptively identify the neighbors of the data objects based on the Davies-Bouldin (DB) index, so it can choose the best clustering results. Experiments indicate that the proposed method may take more time to run, but the algorithm have advantages in detecting the amount and the accuracy of clusters.