Title: Model-based learning via mixtures of contaminated shifted asymmetric Laplace distributions
Authors: Brian Franczak - MacEwan University (Canada) [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 discuss a mixture of shifted asymmetric Laplace (SAL) distributions and extensions thereof. Specifically, we will focus on the development of mixtures of contaminated shifted asymmetric Laplace factor analyzers (MCSALFA). Compared to the well-known mixtures of Gaussian distributions, the mixtures of SAL distributions can parameterize skewness in addition to both location and scale, making it well suited for the analysis of data with homogeneous subpopulations that are not symmetric. In addition to providing a classification of similar observations, the MCSALFA will also provide a classification of an observation as being either `good' or `bad', unifying the fields of classification and outlier detection. Details regarding the development of the MCSALFA will be provided and an alternating-expectation conditional-maximization based parameter estimation scheme will be outlined. The classification performance of these mixtures will be demonstrated using several real data sets.