Title: On model-based learning and directional outlier detection
Authors: Cristina Tortora - San Jose State University (United States) [presenting]
Abstract: Classification can be lucidly defined as the process of assigning group labels to sets of observations. When a finite mixture model is used for classification in either an unsupervised, semi-supervised, or supervised setting, one can refer to this process as model-based learning. We will present a paradigm for parameterizing contamination and skewness within variants of the mixtures of shifted asymmetric Laplace (SAL) distributions. These models will be able to provide both group labels for like observations and detect whether an observation is an outlying point, unifying the fields of model-based learning and outlier detection. Of particular interest are the multiple scaled variants of the mixtures of SAL distributions which allow for directional contamination and skewness, resulting in contours that do not have the traditional elliptical shapes. Explicit details regarding the development of the proposed models will be provided and an expectation-maximization based parameter estimation scheme will be outlined. The classification performance of these models will be demonstrated using simulated and real data sets.