科研之友
微信
新浪微博
Facebook
分享链接
摘要
Clustering is an essential approach for detecting the intrinsic groups in data. An efficient clustering algorithm based on a generalized local synchronization model is proposed. It uses a novel stopping criterion of data synchronization to detect clusters prior to the perfect synchronization. Moreover, a density-biased sampling method is adopted to extract samples from the original data set. The clustering structure can be effectively revealed on the samples. As a result, the clustering efficiency is significantly improved. By using a cluster validity criterion, the proposed algorithm can find clusters of arbitrary number, shape, size and density as well as isolate noises in the vector data without any data distribution assumption. Extensive experiments on several synthetic and real-world data sets show that the proposed algorithm possesses high accuracy and it is more efficient than the state-of-the-art synchronization-based clustering method.
