Title :
Scaling Up Synchronization-Inspired Partitioning Clustering
Author :
Wenhao Ying ; Fu-Lai Chung ; Shitong Wang
Author_Institution :
Sch. of Digital Media, Jiangnan Univ., Wuxi, China
Abstract :
Based on the extensive Kuramoto model, synchronization-inspired partitioning clustering algorithm was recently proposed and is attracting more and more attentions, due to the fact that it simulates the synchronization phenomena in clustering where each data object is regarded as a phase oscillator and the dynamic behavior of the objects is simulated over time. In order to circumvent the serious difficulty that its existing version can only be effectively carried out on considerably small/medium datasets, a novel scalable synchronization-inspired partitioning clustering algorithm termed LSSPC, based on the center-constrained minimal enclosing ball and the reduced set density estimator, is proposed for large dataset applications. LSSPC first condenses a large scale dataset into its reduced dataset by using a fast minimal-enclosing-ball based approximation for the reduced set density estimator, thus achieving an asymptotic time complexity that is linear in the size of dataset and a space complexity that is independent of this size. Then it carries out clustering adaptively on the obtained reduced dataset by using Sync with the Davies-Bouldin clustering criterion and a new order parameter which can help us observe the degree of local synchronization. Finally, it finishes clustering by using the proposed algorithm CRD on the remaining objects in the large dataset, which can capture the outliers and isolated clusters effectively. The effectiveness of the proposed clustering algorithm LSSPC for large datasets is theoretically analyzed and experimentally verified by running on artificial and real datasets.
Keywords :
computational complexity; pattern clustering; synchronisation; CRD algorithm; Davies-Bouldin clustering criterion; Kuramoto model; LSSPC algorithm; Sync; artificial datasets; asymptotic time complexity; center-constrained minimal-enclosing-ball based approximation; dataset size; dynamic object behavior simulation; isolated cluster capturing; large-scale dataset applications; local synchronization degree; order parameter; outlier capturing; phase oscillator; real datasets; reduced set density estimator; scalable synchronization-inspired partitioning clustering algorithm; space complexity; Approximation algorithms; Approximation methods; Clustering algorithms; Heuristic algorithms; Oscillators; Partitioning algorithms; Synchronization; KDE based density estimation; large datasets; minimal enclosing ball; reduced set; synchronization-inspired partitioning clustering;
Journal_Title :
Knowledge and Data Engineering, IEEE Transactions on
DOI :
10.1109/TKDE.2013.178