Title: New algorithms for wrapped normal models estimation
Authors: Anahita Nodehi - Tarbiat Modares University (Iran) [presenting]
Claudio Agostinelli - University of Trento (Italy)
Mousa Golalizadeh - Tarbiat Modares University (Iran)
Abstract: There are a lot of discussions in every statistical context about how to estimate the parameters after choosing a model. One of the crucial problem which deal with wrapped normal distribution is estimating the parameters, especially in multivariate cases. This is due to the form of the density function which is constituted by large sums, and cannot be simplified as close form. The likelihood-based inference for such distribution can be very complicated and computationally intensive. Also, periodic feature of data makes all methods in hands infeasible. The statistics to deal with such data is called directional statistics. Since the shortest distance between two points are not straight line as Euclidean space, it is worth to extend existing methods for such data. Two fast and reliable methods based on Expectation-Maximization (EM) and Classification Expectation-Maximization (CEM) algorithm are suggested. We show the performance of proposal methods in simulation study and real application in compare to existing iterative method.