Density Peaks Clustering for Complex Datasets

Abstract

Clustering by fast search and find of density peaks (DP) is a new density based clustering method and has gained much popularity among the researcher. DP provided the new insight to detect cluster centers and noise in the dataset. DP reveals that a cluster center is a point that have higher density as compared with its neighbor points and have a large distance from other higher density peak points. DP detects each density peak in dataset and discover cluster center with the help of decision graph with minimum human interpretation. After successful identification of cluster centers rest of points are assigned to each cluster center based on the minimum nearest neighbor. DP works very well when each cluster consists of single density however, for more complex and density connected clusters it cannot finds the accurate clusters. To make DP effective equally for more complex datasets, we introduce a novel approach to detect miss classified density and then assign separate density to appropriate cluster. To evaluate the robustness of proposed method we utilized three complex synthetic datasets and compared with DP.