|
Abstract:
|
The classical multivariate analysis influenced by outlying points in the data. Many methods that are resistant to the outliers have been studied in the literature. When the number of outlying objects become larger and have similar or different type of distribution from the major part of the data, they can be considered as another component from the mixture distribution and in this case parameter estimates can be compared with the proposed robust methods. Besides, when the number of outlying objects is smaller, we should be able to detect the major part of the data. Finding highly efficient estimates when there is no data contamination and at the same time, high resistance to outliers, i.e. provide lower bias is not always an easy task. In this study, we plan to perform a comparison study of robustness based on minimum integrated square error estimation (L2E) including the partial L2E, with the well-known multivariate covariance and location estimates; Minimum Covariance Determinant (MCD), Fast algorithm for Minimum Covariance Determinant (FastMCD), and Minimum Volume Ellipsoid(MVE). We test the performance multiple simulation study with different types of data contamination.
|
