XIE Juan-ying1,2, GUO Wen-juan1, XIE Wei-xin2,3
(1 College of Computer Science, Shaanxi Normal University, Xi′an 710062, Shaanxi, China; 2 College of Electronic Engineering, Xidian University, Xi′an 710071, Shaanxi, China; 3 National Laboratory of Automatic Target Recognition (ATR), College of Information Engineering, Shenzhen University, Shenzhen 518060, Guangdong, China)
Abstract:
A new K-medoids algorithm is proposed based on the neighborhood of samples in a dataset. This algorithm defines the radius of the neighborhood and the related neighborhood for each sample according to the distribution of samples in a dataset, and selects the samples that not only lie in higher density area, but also are far away from each other as initial seeds for K-medoids algorithm to overcome the potential disadvantage of the fast K-medoids algorithm to select the samples in the same cluster as initial seeds for different clusters. The proposed K-medoids algorithm is tested on some well-known data sets from UCI machine learning repository and on some synthetic datasets with noisy samples. The experimental results demonstrate that the proposed algorithm achieves the excellent clustering result in short time and is not sensible to noisy data. It outperforms the traditional K-medoids algorithm(Partitioning Around Medoids,PAM) and the fast K-medoids algorithm of Park′s.
KeyWords:
neighborhood; K-medoids clustering algorithm; sample density; clustering; sample space distribution
