Abstract:  In response to the shortcomings of the density peak clustering algorithm, a novel algorithm named adaptive density peak clustering based on comparative parameters is proposed. In the proposed algorithm, a new metric named relative local density was used to assess the similarity between clusters, which greatly improved its applicability to datasets. Another variable called relative connectivity distance was applied for measuring the similarity between clusters, which effectively eliminates the influence of different sizes of clusters in the dataset. The applicability of the algorithm on different datasets was enhanced. By constructing a comentropy function, the parameters could be adaptively determined according to the characteristics of the datasets, which improved the intelligence of the algorithm. A new point allocation strategy was proposed to avoid the effects of the ‘chain reaction’. Experimental results show that the proposed algorithm significantly improves clustering performance compared to the standard and improved density peak clustering algorithm.

Key words: clustering, density peak, adaptive density peak clustering, local density, connectivity distance