Friday, Feb 23 2:00-3:00 in NS 333

Shuchismita Sarkar

Ph.D. candidate at the University of Alabama, Tuscaloosa

Title: Use of transformation in finite mixture modeling and model-based clustering

 Abstract: Finite mixtures present a powerful tool for modeling complex heterogeneous data. One of their most important applications is model-based clustering. It assumes that each data group can be reasonably described by one of mixture model components. This establishes a one-to-one relationship between mixture components and clusters. While the mixture components can take any functional form, the assumption of Gaussian density is one of the most popular in literature. Gaussian mixtures, however, are not efficient when data deviate from normality. Some novel use of transformations for alleviating this issue will be discussed in the case of inconsistent recording of data due to the use of different scales, operator errors, or simply various recording styles. The idea will be extended in analyzing a skewed tensor variate data set consisting of self-reported salaries of teaching faculty employed in American institutions grouped by gender and faculty rank over thirteen years