Title: Outlier detection with mixtures of Gaussian and heavy-tailed distributions
Authors: Alexandra Posekany - University of Technology Vienna (Austria) [presenting]
Abstract: Linear models generally assume normality, a prerequisite often disregarded by data in fields like biology or economics. Our primary aim is to robustify Bayesian inference with mixture models which simultaneously allows for density estimation and outlier detection. To this end, we suggest mixing Students $t$ distributed components in addition to Gaussian ones for identifying the over-dispersed part of data a part of which is extremely noisy, while the rest is normally distributed. To this effect, we employ microarray data as a case study for this behaviour, as they are well-known for their complicated, over-dispersed noise behaviour. Our secondary goal is to present a methodology, which helps not only to identify noisy genes but also to recognise whether single arrays are responsible for this behaviour.