A Multi-Density Clustering Algorithm Based on Similarity for Dataset With Density Variation

Clustering has been widely used in the fields of knowledge discovery, pattern recognition and artificial intelligence. However, discovering clusters in spatial databases is still a challenging task, especially when the shape, size, and density of clusters vary a lot. Existing algorithms have sensitive parameters, clusters must be separated far enough from each other and rich prior knowledge about datasets is required. In this paper, we propose algorithm DENSS, which performs clustering on the basis of the similarity of neighbour distribution and the number of shared neighbors for two objects. Algorithm DENSS can mine clusters that differ in densities, and within a cluster the local densities are reasonably homogeneous. Adjacent objects are separated into different clusters by significant change in densities. To verify the effectiveness of the algorithm DENSS, synthetic and real-world datasets are used for testing, and it has been compared with seven clustering algorithms. Experimental results show that the proposed algorithm has a relatively high efficiency, robustness and effectiveness, and is remarkably superior to the seven algorithms. This algorithm is universal and can rapidly and efficiently identify the clusters of different densities, shapes and sizes even in the presence of noise and outliers for any object feature types.