Possiblistic C-Means Clustering on Directional Data

Statistical analysis of directional data has a long history in the literature where probability distributions, such as von Mises, Fisher, wrapped Cauchy, and wrapped normal, have been used to model data that indicates directional orientations. This paper aims to analyze these data through clustering algorithms. Most of clustering algorithms are based on mixture distribution models or extensions of the k-means algorithm, such as fuzzy c-means (FCM) and possibilistic c-means (PCM). In this paper, we propose PCM direction (PCM-D) algorithm for clustering directional data which is not necessarily based on mixture distribution models by extending PCM to directional data. The proposed PCM-D can perform well for circular, spherical and even hyperspherical data. Comparisons are made with some existing directional clustering algorithms through simulations to show its efficiency. PCM-D was also applied in clustering real directional data.