A1465
Title: On the use of contaminated Gaussian distributions for modeling heavy tails and outliers
Authors: Yana Melnykov - The University of Alabama (United States) [presenting]
Abstract: Gaussian mixture modeling is popular among researchers and practitioners due to its interpretability and relatively straightforward mathematical handling. However, using Gaussian mixtures can be problematic when dealing with outliers and heavy-tailed data groups. The existing literature offers several methods to tackle these issues, with one prominent approach involving the use of contaminated normal distributions. These distributions represent a mixture of two normal components with a common location parameter and one scale parameter being the multiple of the other one. This model allows improved capturing of the potentially heavy distribution tails. Another popular use of contaminated normal distributions is to detect mild outliers. The analysis of both applications of contaminated normal distributions is considered, providing novel insights into the use of this useful model.
