An Improved Density Peaks Method for Data Clustering

Clustering is a powerful approach for data analysis and its aim is to group objects based on their similarities. Density peaks clustering is a recently introduced clustering method with the advantages of doesn't need any predefined parameters and neither any iterative process. In this paper, a novel density peaks clustering method called IDPC is proposed. The proposed method consists of two major steps. In the first step, local density concept is used to identify cluster centers. In the second step, a novel label propagation method is proposed to form clusters. The proposed label propagation method also uses the local density concept in its process to propagate the cluster labels around the whole data points. The effectiveness of the proposed method has been assessed on a synthetic datasets and also on some real-world datasets. The obtained results show that the proposed method outperformed the other state-of-the art methods.