Content driven clustering algorithm combining density and distance functions. (March 2019)

• Record Type: Journal Article
• Authors: Ikonomakis, Emmanouil K.; Spyrou, George M.; Vrahatis, Michael N.
• Is Part Of: Pattern recognition. Volume 87(2019:Mar.)
• Page Start: 190
• Page End: 202
• Publication Date: 2019-03
• Subjects: Clustering algorithms -- Density based clustering -- Distance based clustering -- Evolutionary clustering -- Window density function; Pattern perception -- Periodicals; Perception des structures -- Périodiques; Patroonherkenning; 006.4
• DOI: 10.1016/j.patcog.2018.10.007
• Languages: English
• ISSNs: 0031-3203

Abstract

Highlights: A density and distance based hybrid algorithm is presented. The algorithm is capable of adapting itself based on the size of the dataset. The algorithm can use any evolutionary optimization algorithm. Its performance is compared to AP, k-means and DBSCAN by using the V-measure. It outperforms all other algorithms while having a low deviation in running times. Abstract: Density and distance based clustering are two distinct approaches to the same problem. In this contribution, a novel algorithm is presented in order to exploit the benefits of both approaches. This is achieved, not by combining those approaches into a single notion, but by utilizing the advantages of each one, depending on what each step of the algorithm aims to achieve. To be precise, the Window Density Function is utilized to provide regions of high density and hence a region of clusters or a part of a cluster. Affinity Propagation is, consequently, utilized to provide a group of clusters within such a region. Finally, these regions are merged to form actual clusters. The proposed methodology is tested on a variety of synthetic and real-life datasets. The algorithm presented in this contribution outperforms other well-known algorithms, with which it is compared to, in the majority of the datasets used.