Author:
Chang-Chien Shou-Jen,Ali Wajid,Yang Miin-Shen
Abstract
Clustering is a method for analyzing grouped data. Circular data were well used in various applications, such as wind directions, departure directions of migrating birds or animals, etc. The expectation & maximization (EM) algorithm on mixtures of von Mises distributions is popularly used for clustering circular data. In general, the EM algorithm is sensitive to initials and not robust to outliers in which it is also necessary to give a number of clusters a priori. In this paper, we consider a learning-based schema for EM, and then propose a learning-based EM algorithm on mixtures of von Mises distributions for clustering grouped circular data. The proposed clustering method is without any initial and robust to outliers with automatically finding the number of clusters. Some numerical and real data sets are used to compare the proposed algorithm with existing methods. Experimental results and comparisons actually demonstrate these good aspects of effectiveness and superiority of the proposed learning-based EM algorithm.
Publisher
Taiwan Association of Engineering and Technology Innovation
Reference21 articles.
1. R. Von Mises, “Uber die ‘Ganzzahligkeit’ der atomgewicht und verwandte fragen,” Physikal Z., vol. 19, pp. 490-500, 1918.
2. G. S. Watson and E. J. Williams, “On the construction of significance tests on the circle and the sphere,” Biometrika, vol. 43, no. 3/4, pp. 344-352, December 1956.
3. N. I. Fisher, Statistical analysis of circular data, Cambridge:Cambridge University, in press, October 1995.
4. N. Masseran, A. M. Razali, K. Ibrahim, and M. T. Latif, “Fitting a mixture of von Mises distributions in order to model data on wind direction in Peninsular Malaysia,” Energy Conversion and Management, vol. 72, pp. 94-102, April 2013.
5. L. P. Rivest and S. Kato, “A random‐effects model for clustered circular data,” Canadian Journal of Statistics, vol. 47, no. 4, pp. 712-728, August 2019.
Cited by
1 articles.
订阅此论文施引文献
订阅此论文施引文献,注册后可以免费订阅5篇论文的施引文献,订阅后可以查看论文全部施引文献