Hierarchical clustering is of great importance in data analysis. Although there are a number of hierarchical clustering algorithms including agglomerative methods, divisive methods and hybrid methods, most of them are sensitive to noise points, suffer from high computational cost and cannot effectively discover clusters with complex structures. When recognizing patterns from complex structures, humans intuitively tend to discover obvious clusters in dense regions firstly and then deal with objects on the border. Inspired by this idea, we propose a local cores-based hierarchical clustering algorithm called HCLORE. The proposed method first partitions the data set into several clusters by finding local cores, instead of optimizing an objective function through iteration like K-means; then, temporarily removes points with lower local density, so that the boundary between clusters is clearer; after that, merges clusters according to a new defined similarities between clusters; and finally, points with lower local density are assigned to the same clusters as their local cores belong to. The experimental results on synthetic data sets and real data sets show that our algorithm is more effective and efficient than existing methods when processing data sets with complex structures.
展开▼
