摘要： In this paper, we present a new algorithm for clustering proximity-relation matrix that does not require the transitivity property. The proposed algorithm is first inspired by the idea of Yang and Wu [16] then turned into a self-organizing process that is built upon the intuition behind clustering. At the end of the process subjects belonging to be the same cluster should converge to the same point, which represents the cluster center. However, the performance of Yang and Wu's algorithm depends on parameter selection. In this paper, we use the partition entropy (PE) index to choose it. Numerical result illustrates that the proposed method does not only solve the parameter selection problem but also obtains an optimal clustering result. Finally, we apply the proposed algorithm to three applications. One is to evaluate the performance of higher education in Taiwan, another is machine-parts grouping in cellular manufacturing systems, and the other is to cluster probability density functions. 出版者： Abingdon: Taylor & Francis 出版日期： 2013-09-01 出處： Journal of applied statistics, 2013-09, Vol.40 (9), p.1875-1892 資源來源： Taylor & Francis Journals Auto-Holdings Collection 版權： Copyright Taylor & Francis Group, LLC 2013 版權： Copyright Taylor & Francis Ltd. 2013 識別號： ISSN: 0266-4763 識別號： EISSN: 1360-0532 識別號： DOI: 10.1080/02664763.2013.799126
