Finite mixture modeling based on the scale and shape mixtures of multivariate skew-normal distribution
碩士 === 國立中興大學 === 統計學研究所 === 105 === We introduce a novel class of scale and shape mixtures of multivariate skew-normal distributions that allow for modeling multivariate asymmetric data in a wide range of considerations. We study some of their general properties and theoretic foundations for furthe...
| Main Authors: | , |
|---|---|
| Other Authors: | |
| Format: | Others |
| Language: | zh-TW |
| Published: |
2017
|
| Online Access: | http://ndltd.ncl.edu.tw/handle/90225944978103355748 |
| Summary: | 碩士 === 國立中興大學 === 統計學研究所 === 105 === We introduce a novel class of scale and shape mixtures of multivariate skew-normal distributions that allow for modeling multivariate asymmetric data in a wide range of considerations. We study some of their general properties and theoretic foundations for further inferential investigations. Afterword, we present a framework of mixture modeling based on this family of distributions as a new robust model-based tool which aims to group data containing at least one group of observations with fat tails or possible asymmetric observations. A computational tractable ECM algorithm is developed to conduct maximum likelihood estimation of parameters and the resulting expressions involved in the E- and CM-steps are shown to be analytically simple. The asymptotic covariance matrix of parameter estimates is derived from the observed information matrix using the outer product of expected complete-data scores.
We demonstrate the utility of the proposed approach using two simulated studies and two real-data examples.
|
|---|