WITHDRAWN: On Model-Based Clustering of Directional Data with Heavy Tails and Outliers

• Activity Number: 321 - Machine Learning and Variable Selection
• Type: Contributed
• Date/Time: Wednesday, August 11, 2021 : 3:30 PM to 5:20 PM
• Sponsor: Section on Statistical Computing
• Author(s): Yingying Zhang and Volodymyr Melnykov and Igor Melnykov
• Companies: University of South Alabama and University of Alabama and University of Minnesota Duluth
• Keywords: EM algorithm; directional data; von Mises-Fisher distribution; mixture model

Abstract

Directional statistics deals with data that can be naturally expressed in the form of vector directions. In most cases, such data lie on the surface of the unit sphere. The most challenging aspect of working with directional data is the lack of flexible models, especially in multivariate setting. von Mises-Fisher distribution is one of the most fundamental and commonly used parametric models to describe directional data. Mixture models of von Mises-Fisher distributions have already been developed to hand heterogeneous observations. Unfortunately traditional models are not robust to the presence of noise, outliers, and heavy tails. In this paper, mixture of contaminated von Mises-Fisher distributions is considered as a more appropriate alternative in this setting. The performance of the proposed methodology is tested on synthetic and real-life data. The procedure demonstrates its superiority over the traditional model-based clustering procedure relying on the mixture of von Mises-Fisher distributions.