Optimized adaptive gravitational density peak clustering algorithm

Aiming to address the issues of sensitivity to truncation distance, lack of unified density measurement standard, and subjective selection of cluster center in Density Peak Clustering algorithm (DPC) , we propose an Optimized Adaptive Gravitational Density Peak Clustering algorithm (OAGDPC). Firstly, we employ the Fuzzy Weighted K-Nearest Neighbors Density Peak Clustering (FKNN-DPC) to redefine local density and establish a unified density measurement standard. Then, an adaptive strategy for selecting cluster centers is introduced. This is achieved by incorporating the parameter mapping between Newton's law of gravity and the Gravitational Density Peak Clustering (GDPC) algorithm, utilizing the gravitational analogy distance and considering both local density and gravity in decision parameter settings, it achieves adaptive determination of clustering centers through monitoring changes in the top angle of the descending broken line graph formed by these decision parameters. Subsequently, non-center points are gathered, and abnormal points are identified. The experiments select DPC, GPDC, FKNN-DPC and OAGDPC to test on artificial and UCI data sets. The results show that OAGDPC algorithm has good performance on all data sets, especially in the accuracy, adaptive ability and robustness of the clustering results compared with the comparison algorithms.