GDPC: Gravitation-based Density Peaks Clustering algorithm

The Density Peaks Clustering algorithm, which we refer to as DPC, is a novel and efficient density-based clustering approach, and it is published in Science in 2014. The DPC has advantages of discovering clusters with varying sizes and varying densities, but has some limitations of detecting the num...

Full description

Saved in:
Bibliographic Details
Published in:Physica A 2018-07, Vol.502, p.345-355
Main Authors: Jiang, Jianhua, Hao, Dehao, Chen, Yujun, Parmar, Milan, Li, Keqin
Format: Article
Language:English
Subjects:
Anomaly detection
Clustering analysis
Density peaks clustering
Gravitation theory
Citations: Items that this one cites
Items that cite this one
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:The Density Peaks Clustering algorithm, which we refer to as DPC, is a novel and efficient density-based clustering approach, and it is published in Science in 2014. The DPC has advantages of discovering clusters with varying sizes and varying densities, but has some limitations of detecting the number of clusters and identifying anomalies. We develop an enhanced algorithm with an alternative decision graph based on gravitation theory and nearby distance to identify centroids and anomalies accurately. We apply our method to some UCI and synthetic data sets. We report comparative clustering performances using F-Measure and 2-dimensional vision. We also compare our method to other clustering algorithms, such as K-Means, Affinity Propagation (AP) and DPC. We present F-Measure scores and clustering accuracies of our GDPC algorithm compared to K-Means, AP and DPC on different data sets. We show that the GDPC has the superior performance in its capability of: (1) detecting the number of clusters obviously; (2) aggregating clusters with varying sizes, varying densities efficiently; (3) identifying anomalies accurately.
ISSN:0378-4371
1873-2119
DOI:10.1016/j.physa.2018.02.084