关键词: 聚类, 局部密度, 密度峰值, K近邻, 多类合并

Abstract: Aiming at the problem that the density peaks clustering (DPC) algorithm had poor clustering performance in dealing with data with complex structure, high dimensionality and multiple density peaks in the same class, we proposed a density peaks clustering algorithm based on Knearest neighbors and classesmerging (KMDPC). Firstly, the sample distribution was described by the defined density calculation method, and the clustering center was obtained by using new evalution index. Secondly, an iterative assignment strategy based on the idea of Knearest neighbors was designed to classify the remaining data points accurately. Finally, a local merging method was presented to prevent the splitting of classes with mu
ltiple density peaks. Simulation results show that the performance of this algorithm is obviously better than that of DPC algorithm on 22 different datasets.

Key words: clustering, local density, density peak, K-nearest , neighbor, classesmerging