A New Method of Eliminating Noise Based on Clustering

Clustering is often constructed on noise-free datasets. In real-world applications, it is inevitable that the datasets contain noises, which may result in unsatisfactory results of the clustering algorithms. In this paper, several methods of reducing noises are systemic introduced, and at the first time we propose a heuristic algorithm of reducing noises in clustering theory (GK-means). The empirical results show that GK-means is simpler and more precise, and can handle noises in the real-world database effectively. Some samples are used to prove the validity of this algorithm.