Title: An improved method for density-based clustering

Authors: Hong Jin; Shuliang Wang; Qian Zhou; Ying Li

Addresses: State Key Laboratory of Software Engineering, Wuhan University, Wuhan 430079, China ' International School of Software, State Key Laboratory of Information Engineering in Surveying, Mapping and Remote Sensing, Wuhan University, Wuhan 430079, China ' International School of Software, Wuhan University, Wuhan 430079, China ' State Key Laboratory of Information Engineering in Surveying, Mapping and Remote Sensing, Wuhan University, Wuhan 430079, China

Abstract: Knowledge discovery in large multimedia databases which usually contain large amounts of noise and high-dimensional feature vectors is an increasingly important research issue. Density-based clustering is proved to be much more efficient when dealing with such databases. However, its clustering quality mainly depends on the parameter setting. For the adequate choice of the parameters to be preset, it has difficulty in its operability without enough domain knowledge. To solve such problem, in this paper it proposed a new approach to immediately inference an appropriate value for one of the parameters named bandwidth. Based on the Bayesian Theorem, it is to infer the suitable parameter value by the constructed parameter estimation model. Then the user only has to preset the other parameter noise threshold. As a result, the clusters can be identified by the determined parameter values. The experimental results show that the proposed method has complementary advantages in the density-based clustering algorithm.

Keywords: density-based clustering; DENCLUE; optimal bandwidth selection; Bayesian posterior probability estimation; knowledge discovery; multimedia databases; parameter estimation.

DOI: 10.1504/IJDMMM.2014.066763

International Journal of Data Mining, Modelling and Management, 2014 Vol.6 No.4, pp.347 - 368

Received: 08 May 2021
Accepted: 12 May 2021

Published online: 03 Jan 2015 *

Full-text access for editors Access for subscribers Purchase this article Comment on this article