A novel density deviation multi-peaks automatic clustering algorithm

Abstract The density peaks clustering (DPC) algorithm is a classical and widely used method. However, the DPC requires manual selection of cluster centers, single way calculation, cannot effectively handle low-density points. To address above issues, we propose novel deviation multi-peaks automatic method (AmDPC) in this paper. Firstly, new local-density use to measure relationship between data points cut-off distance ( $$d_c$$ d c ). Secondly, divide into multiple levels equally extract with higher distances each level. Finally, for multi-peak at levels, merge them according size difference deviation. We finally achieve overall by processing verify performance method, test synthetic dataset, real-world Olivetti Face respectively. simulation experimental results indicate that AmDPC can more has certain effectiveness robustness.