A0794
Title: Recent developments in using mixtures of multivariate asymmetric distributions for classification
Authors: Brian Franczak - MacEwan University (Canada) [presenting]
Abstract: Classification is defined as the process of assigning group labels to unlabelled observations. Classification is performed in multiple ways; for example, one can utilize unsupervised, semi-supervised, or fully supervised techniques when a finite mixture model is used for classification, called process model-based classification. Recent advances in the development of mixtures with multivariate density function capable of modelling skewness directly are discussed. In the context of classification applications, topics may include one or more strategies for handling data with missing values, working with high-dimensional data sets, rectifying issues with typical parameter estimation schemes, parameterizing tail-weight separately in each dimension of the data, or accounting for outlying or spurious observations. The proposed models will be demonstrated using both simulated and real data. Model performance will be assessed using standard metrics and by comparison to popular publicly available methods.
