A0181
Title: High-dimensional LDA biplot through the GSVD
Authors: Raeesa Ganey - University of Witwatersrand (South Africa) [presenting]
Abstract: Linear discriminant Analysis is a multivariate technique concerned with separating distinct sets of observations. However, a common limitation of trace optimisation in discriminant analysis is that the within-cluster scatter matrix must be nonsingular, which restricts the use of data sets when the number of variables is larger than the number of observations. The same goal of discriminant analysis can be achieved by applying the generalised singular value decomposition (GSVD) regardless of the number of variables. We present that by using this approach, we can easily apply discriminant analysis and construct graphical representations to such data. We will look at Canonical Variate Analysis (CVA) biplots that will display observations as points and variables as axes in a reduced dimension, providing a highly informative visual display of the respective class separations.
