Multilevel Core-sets Based Aggregation Clustering Algorithm

As an important component of unsupervised pattern classification, clustering analysis has becomean more and more important tool in modern data analysis. Different starting conditions and similaritycriterion and clustering criterion usually generate different clustering algorithms. Therefore, clusteringalgorithm is a vast and diverse algorithm family. So far, many researchers has put forward variousclustering algorithms.Many classical clustering algorithms like average-link, K-means, K-medoids, Clarans, Clara and soon are all based on a single cluster-center and only apt to discover convex-structured clusters. Othermethods, e.g., CURE and DBSCAN, use more than one point to represent a cluster and can find somewell-separated clusters of arbitrary shape. However, they only think the original scale of the inputdata and thus cannot depart over-lapped or noisy clusters. To this end, this paper is to propose amultilevel core-set based agglomerative clustering algorithm (MulCA). The idea of MulCA is that theclustering structure is described by "multi-layer core set". Clustering Process is achieved throughprocedure which the top of the "core set" automatically becomes the underlying data set.As the rapidly development of modern society, large-scale data sets have become the main memberof datasets to be dealt with. For that, through the introduction of "random sampling based ε-core set(RBC)", MulCA algorithm is applied to large-scale data sets. A large number of numericalexperiments fully verify the algorithm MulCA.The experimental results on both synthetic and real-world datasets have demonstrated theeffectiveness of the algorithm MulCA, and the superiority of it compared to other algorithms.